2 * Generic process-grouping system.
4 * Based originally on the cpuset system, extracted by Paul Menage
5 * Copyright (C) 2006 Google, Inc
7 * Notifications support
8 * Copyright (C) 2009 Nokia Corporation
9 * Author: Kirill A. Shutemov
11 * Copyright notices from the original cpuset code:
12 * --------------------------------------------------
13 * Copyright (C) 2003 BULL SA.
14 * Copyright (C) 2004-2006 Silicon Graphics, Inc.
16 * Portions derived from Patrick Mochel's sysfs code.
17 * sysfs is Copyright (c) 2001-3 Patrick Mochel
19 * 2003-10-10 Written by Simon Derr.
20 * 2003-10-22 Updates by Stephen Hemminger.
21 * 2004 May-July Rework by Paul Jackson.
22 * ---------------------------------------------------
24 * This file is subject to the terms and conditions of the GNU General Public
25 * License. See the file COPYING in the main directory of the Linux
26 * distribution for more details.
29 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
31 #include "cgroup-internal.h"
33 #include <linux/cred.h>
34 #include <linux/errno.h>
35 #include <linux/init_task.h>
36 #include <linux/kernel.h>
37 #include <linux/magic.h>
38 #include <linux/mutex.h>
39 #include <linux/mount.h>
40 #include <linux/pagemap.h>
41 #include <linux/proc_fs.h>
42 #include <linux/rcupdate.h>
43 #include <linux/sched.h>
44 #include <linux/sched/task.h>
45 #include <linux/slab.h>
46 #include <linux/spinlock.h>
47 #include <linux/percpu-rwsem.h>
48 #include <linux/string.h>
49 #include <linux/hashtable.h>
50 #include <linux/idr.h>
51 #include <linux/kthread.h>
52 #include <linux/atomic.h>
53 #include <linux/cpuset.h>
54 #include <linux/proc_ns.h>
55 #include <linux/nsproxy.h>
56 #include <linux/file.h>
59 #define CREATE_TRACE_POINTS
60 #include <trace/events/cgroup.h>
62 #define CGROUP_FILE_NAME_MAX (MAX_CGROUP_TYPE_NAMELEN + \
66 * cgroup_mutex is the master lock. Any modification to cgroup or its
67 * hierarchy must be performed while holding it.
69 * css_set_lock protects task->cgroups pointer, the list of css_set
70 * objects, and the chain of tasks off each css_set.
72 * These locks are exported if CONFIG_PROVE_RCU so that accessors in
73 * cgroup.h can use them for lockdep annotations.
75 DEFINE_MUTEX(cgroup_mutex);
76 DEFINE_SPINLOCK(css_set_lock);
78 #ifdef CONFIG_PROVE_RCU
79 EXPORT_SYMBOL_GPL(cgroup_mutex);
80 EXPORT_SYMBOL_GPL(css_set_lock);
84 * Protects cgroup_idr and css_idr so that IDs can be released without
85 * grabbing cgroup_mutex.
87 static DEFINE_SPINLOCK(cgroup_idr_lock);
90 * Protects cgroup_file->kn for !self csses. It synchronizes notifications
91 * against file removal/re-creation across css hiding.
93 static DEFINE_SPINLOCK(cgroup_file_kn_lock);
95 struct percpu_rw_semaphore cgroup_threadgroup_rwsem;
97 #define cgroup_assert_mutex_or_rcu_locked() \
98 RCU_LOCKDEP_WARN(!rcu_read_lock_held() && \
99 !lockdep_is_held(&cgroup_mutex), \
100 "cgroup_mutex or RCU read lock required");
103 * cgroup destruction makes heavy use of work items and there can be a lot
104 * of concurrent destructions. Use a separate workqueue so that cgroup
105 * destruction work items don't end up filling up max_active of system_wq
106 * which may lead to deadlock.
108 static struct workqueue_struct *cgroup_destroy_wq;
110 /* generate an array of cgroup subsystem pointers */
111 #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys,
112 struct cgroup_subsys *cgroup_subsys[] = {
113 #include <linux/cgroup_subsys.h>
117 /* array of cgroup subsystem names */
118 #define SUBSYS(_x) [_x ## _cgrp_id] = #_x,
119 static const char *cgroup_subsys_name[] = {
120 #include <linux/cgroup_subsys.h>
124 /* array of static_keys for cgroup_subsys_enabled() and cgroup_subsys_on_dfl() */
126 DEFINE_STATIC_KEY_TRUE(_x ## _cgrp_subsys_enabled_key); \
127 DEFINE_STATIC_KEY_TRUE(_x ## _cgrp_subsys_on_dfl_key); \
128 EXPORT_SYMBOL_GPL(_x ## _cgrp_subsys_enabled_key); \
129 EXPORT_SYMBOL_GPL(_x ## _cgrp_subsys_on_dfl_key);
130 #include <linux/cgroup_subsys.h>
133 #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys_enabled_key,
134 static struct static_key_true *cgroup_subsys_enabled_key[] = {
135 #include <linux/cgroup_subsys.h>
139 #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys_on_dfl_key,
140 static struct static_key_true *cgroup_subsys_on_dfl_key[] = {
141 #include <linux/cgroup_subsys.h>
146 * The default hierarchy, reserved for the subsystems that are otherwise
147 * unattached - it never has more than a single cgroup, and all tasks are
148 * part of that cgroup.
150 struct cgroup_root cgrp_dfl_root;
151 EXPORT_SYMBOL_GPL(cgrp_dfl_root);
154 * The default hierarchy always exists but is hidden until mounted for the
155 * first time. This is for backward compatibility.
157 static bool cgrp_dfl_visible;
159 /* some controllers are not supported in the default hierarchy */
160 static u16 cgrp_dfl_inhibit_ss_mask;
162 /* some controllers are implicitly enabled on the default hierarchy */
163 static u16 cgrp_dfl_implicit_ss_mask;
165 /* some controllers can be threaded on the default hierarchy */
166 static u16 cgrp_dfl_threaded_ss_mask;
168 /* The list of hierarchy roots */
169 LIST_HEAD(cgroup_roots);
170 static int cgroup_root_count;
172 /* hierarchy ID allocation and mapping, protected by cgroup_mutex */
173 static DEFINE_IDR(cgroup_hierarchy_idr);
176 * Assign a monotonically increasing serial number to csses. It guarantees
177 * cgroups with bigger numbers are newer than those with smaller numbers.
178 * Also, as csses are always appended to the parent's ->children list, it
179 * guarantees that sibling csses are always sorted in the ascending serial
180 * number order on the list. Protected by cgroup_mutex.
182 static u64 css_serial_nr_next = 1;
185 * These bitmasks identify subsystems with specific features to avoid
186 * having to do iterative checks repeatedly.
188 static u16 have_fork_callback __read_mostly;
189 static u16 have_exit_callback __read_mostly;
190 static u16 have_release_callback __read_mostly;
191 static u16 have_canfork_callback __read_mostly;
193 /* cgroup namespace for init task */
194 struct cgroup_namespace init_cgroup_ns = {
195 .count = REFCOUNT_INIT(2),
196 .user_ns = &init_user_ns,
197 .ns.ops = &cgroupns_operations,
198 .ns.inum = PROC_CGROUP_INIT_INO,
199 .root_cset = &init_css_set,
202 static struct file_system_type cgroup2_fs_type;
203 static struct cftype cgroup_base_files[];
205 static int cgroup_apply_control(struct cgroup *cgrp);
206 static void cgroup_finalize_control(struct cgroup *cgrp, int ret);
207 static void css_task_iter_skip(struct css_task_iter *it,
208 struct task_struct *task);
209 static int cgroup_destroy_locked(struct cgroup *cgrp);
210 static struct cgroup_subsys_state *css_create(struct cgroup *cgrp,
211 struct cgroup_subsys *ss);
212 static void css_release(struct percpu_ref *ref);
213 static void kill_css(struct cgroup_subsys_state *css);
214 static int cgroup_addrm_files(struct cgroup_subsys_state *css,
215 struct cgroup *cgrp, struct cftype cfts[],
219 * cgroup_ssid_enabled - cgroup subsys enabled test by subsys ID
220 * @ssid: subsys ID of interest
222 * cgroup_subsys_enabled() can only be used with literal subsys names which
223 * is fine for individual subsystems but unsuitable for cgroup core. This
224 * is slower static_key_enabled() based test indexed by @ssid.
226 bool cgroup_ssid_enabled(int ssid)
228 if (CGROUP_SUBSYS_COUNT == 0)
231 return static_key_enabled(cgroup_subsys_enabled_key[ssid]);
235 * cgroup_on_dfl - test whether a cgroup is on the default hierarchy
236 * @cgrp: the cgroup of interest
238 * The default hierarchy is the v2 interface of cgroup and this function
239 * can be used to test whether a cgroup is on the default hierarchy for
240 * cases where a subsystem should behave differnetly depending on the
243 * The set of behaviors which change on the default hierarchy are still
244 * being determined and the mount option is prefixed with __DEVEL__.
246 * List of changed behaviors:
248 * - Mount options "noprefix", "xattr", "clone_children", "release_agent"
249 * and "name" are disallowed.
251 * - When mounting an existing superblock, mount options should match.
253 * - Remount is disallowed.
255 * - rename(2) is disallowed.
257 * - "tasks" is removed. Everything should be at process granularity. Use
258 * "cgroup.procs" instead.
260 * - "cgroup.procs" is not sorted. pids will be unique unless they got
261 * recycled inbetween reads.
263 * - "release_agent" and "notify_on_release" are removed. Replacement
264 * notification mechanism will be implemented.
266 * - "cgroup.clone_children" is removed.
268 * - "cgroup.subtree_populated" is available. Its value is 0 if the cgroup
269 * and its descendants contain no task; otherwise, 1. The file also
270 * generates kernfs notification which can be monitored through poll and
271 * [di]notify when the value of the file changes.
273 * - cpuset: tasks will be kept in empty cpusets when hotplug happens and
274 * take masks of ancestors with non-empty cpus/mems, instead of being
275 * moved to an ancestor.
277 * - cpuset: a task can be moved into an empty cpuset, and again it takes
278 * masks of ancestors.
280 * - memcg: use_hierarchy is on by default and the cgroup file for the flag
283 * - blkcg: blk-throttle becomes properly hierarchical.
285 * - debug: disallowed on the default hierarchy.
287 bool cgroup_on_dfl(const struct cgroup *cgrp)
289 return cgrp->root == &cgrp_dfl_root;
292 /* IDR wrappers which synchronize using cgroup_idr_lock */
293 static int cgroup_idr_alloc(struct idr *idr, void *ptr, int start, int end,
298 idr_preload(gfp_mask);
299 spin_lock_bh(&cgroup_idr_lock);
300 ret = idr_alloc(idr, ptr, start, end, gfp_mask & ~__GFP_DIRECT_RECLAIM);
301 spin_unlock_bh(&cgroup_idr_lock);
306 static void *cgroup_idr_replace(struct idr *idr, void *ptr, int id)
310 spin_lock_bh(&cgroup_idr_lock);
311 ret = idr_replace(idr, ptr, id);
312 spin_unlock_bh(&cgroup_idr_lock);
316 static void cgroup_idr_remove(struct idr *idr, int id)
318 spin_lock_bh(&cgroup_idr_lock);
320 spin_unlock_bh(&cgroup_idr_lock);
323 static bool cgroup_has_tasks(struct cgroup *cgrp)
325 return cgrp->nr_populated_csets;
328 bool cgroup_is_threaded(struct cgroup *cgrp)
330 return cgrp->dom_cgrp != cgrp;
333 /* can @cgrp host both domain and threaded children? */
334 static bool cgroup_is_mixable(struct cgroup *cgrp)
337 * Root isn't under domain level resource control exempting it from
338 * the no-internal-process constraint, so it can serve as a thread
339 * root and a parent of resource domains at the same time.
341 return !cgroup_parent(cgrp);
344 /* can @cgrp become a thread root? should always be true for a thread root */
345 static bool cgroup_can_be_thread_root(struct cgroup *cgrp)
347 /* mixables don't care */
348 if (cgroup_is_mixable(cgrp))
351 /* domain roots can't be nested under threaded */
352 if (cgroup_is_threaded(cgrp))
355 /* can only have either domain or threaded children */
356 if (cgrp->nr_populated_domain_children)
359 /* and no domain controllers can be enabled */
360 if (cgrp->subtree_control & ~cgrp_dfl_threaded_ss_mask)
366 /* is @cgrp root of a threaded subtree? */
367 bool cgroup_is_thread_root(struct cgroup *cgrp)
369 /* thread root should be a domain */
370 if (cgroup_is_threaded(cgrp))
373 /* a domain w/ threaded children is a thread root */
374 if (cgrp->nr_threaded_children)
378 * A domain which has tasks and explicit threaded controllers
379 * enabled is a thread root.
381 if (cgroup_has_tasks(cgrp) &&
382 (cgrp->subtree_control & cgrp_dfl_threaded_ss_mask))
388 /* a domain which isn't connected to the root w/o brekage can't be used */
389 static bool cgroup_is_valid_domain(struct cgroup *cgrp)
391 /* the cgroup itself can be a thread root */
392 if (cgroup_is_threaded(cgrp))
395 /* but the ancestors can't be unless mixable */
396 while ((cgrp = cgroup_parent(cgrp))) {
397 if (!cgroup_is_mixable(cgrp) && cgroup_is_thread_root(cgrp))
399 if (cgroup_is_threaded(cgrp))
406 /* subsystems visibly enabled on a cgroup */
407 static u16 cgroup_control(struct cgroup *cgrp)
409 struct cgroup *parent = cgroup_parent(cgrp);
410 u16 root_ss_mask = cgrp->root->subsys_mask;
413 u16 ss_mask = parent->subtree_control;
415 /* threaded cgroups can only have threaded controllers */
416 if (cgroup_is_threaded(cgrp))
417 ss_mask &= cgrp_dfl_threaded_ss_mask;
421 if (cgroup_on_dfl(cgrp))
422 root_ss_mask &= ~(cgrp_dfl_inhibit_ss_mask |
423 cgrp_dfl_implicit_ss_mask);
427 /* subsystems enabled on a cgroup */
428 static u16 cgroup_ss_mask(struct cgroup *cgrp)
430 struct cgroup *parent = cgroup_parent(cgrp);
433 u16 ss_mask = parent->subtree_ss_mask;
435 /* threaded cgroups can only have threaded controllers */
436 if (cgroup_is_threaded(cgrp))
437 ss_mask &= cgrp_dfl_threaded_ss_mask;
441 return cgrp->root->subsys_mask;
445 * cgroup_css - obtain a cgroup's css for the specified subsystem
446 * @cgrp: the cgroup of interest
447 * @ss: the subsystem of interest (%NULL returns @cgrp->self)
449 * Return @cgrp's css (cgroup_subsys_state) associated with @ss. This
450 * function must be called either under cgroup_mutex or rcu_read_lock() and
451 * the caller is responsible for pinning the returned css if it wants to
452 * keep accessing it outside the said locks. This function may return
453 * %NULL if @cgrp doesn't have @subsys_id enabled.
455 static struct cgroup_subsys_state *cgroup_css(struct cgroup *cgrp,
456 struct cgroup_subsys *ss)
459 return rcu_dereference_check(cgrp->subsys[ss->id],
460 lockdep_is_held(&cgroup_mutex));
466 * cgroup_e_css - obtain a cgroup's effective css for the specified subsystem
467 * @cgrp: the cgroup of interest
468 * @ss: the subsystem of interest (%NULL returns @cgrp->self)
470 * Similar to cgroup_css() but returns the effective css, which is defined
471 * as the matching css of the nearest ancestor including self which has @ss
472 * enabled. If @ss is associated with the hierarchy @cgrp is on, this
473 * function is guaranteed to return non-NULL css.
475 static struct cgroup_subsys_state *cgroup_e_css(struct cgroup *cgrp,
476 struct cgroup_subsys *ss)
478 lockdep_assert_held(&cgroup_mutex);
484 * This function is used while updating css associations and thus
485 * can't test the csses directly. Test ss_mask.
487 while (!(cgroup_ss_mask(cgrp) & (1 << ss->id))) {
488 cgrp = cgroup_parent(cgrp);
493 return cgroup_css(cgrp, ss);
497 * cgroup_get_e_css - get a cgroup's effective css for the specified subsystem
498 * @cgrp: the cgroup of interest
499 * @ss: the subsystem of interest
501 * Find and get the effective css of @cgrp for @ss. The effective css is
502 * defined as the matching css of the nearest ancestor including self which
503 * has @ss enabled. If @ss is not mounted on the hierarchy @cgrp is on,
504 * the root css is returned, so this function always returns a valid css.
505 * The returned css must be put using css_put().
507 struct cgroup_subsys_state *cgroup_get_e_css(struct cgroup *cgrp,
508 struct cgroup_subsys *ss)
510 struct cgroup_subsys_state *css;
515 css = cgroup_css(cgrp, ss);
517 if (css && css_tryget_online(css))
519 cgrp = cgroup_parent(cgrp);
522 css = init_css_set.subsys[ss->id];
529 static void cgroup_get_live(struct cgroup *cgrp)
531 WARN_ON_ONCE(cgroup_is_dead(cgrp));
532 css_get(&cgrp->self);
535 struct cgroup_subsys_state *of_css(struct kernfs_open_file *of)
537 struct cgroup *cgrp = of->kn->parent->priv;
538 struct cftype *cft = of_cft(of);
541 * This is open and unprotected implementation of cgroup_css().
542 * seq_css() is only called from a kernfs file operation which has
543 * an active reference on the file. Because all the subsystem
544 * files are drained before a css is disassociated with a cgroup,
545 * the matching css from the cgroup's subsys table is guaranteed to
546 * be and stay valid until the enclosing operation is complete.
549 return rcu_dereference_raw(cgrp->subsys[cft->ss->id]);
553 EXPORT_SYMBOL_GPL(of_css);
556 * for_each_css - iterate all css's of a cgroup
557 * @css: the iteration cursor
558 * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end
559 * @cgrp: the target cgroup to iterate css's of
561 * Should be called under cgroup_[tree_]mutex.
563 #define for_each_css(css, ssid, cgrp) \
564 for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \
565 if (!((css) = rcu_dereference_check( \
566 (cgrp)->subsys[(ssid)], \
567 lockdep_is_held(&cgroup_mutex)))) { } \
571 * for_each_e_css - iterate all effective css's of a cgroup
572 * @css: the iteration cursor
573 * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end
574 * @cgrp: the target cgroup to iterate css's of
576 * Should be called under cgroup_[tree_]mutex.
578 #define for_each_e_css(css, ssid, cgrp) \
579 for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \
580 if (!((css) = cgroup_e_css(cgrp, cgroup_subsys[(ssid)]))) \
585 * do_each_subsys_mask - filter for_each_subsys with a bitmask
586 * @ss: the iteration cursor
587 * @ssid: the index of @ss, CGROUP_SUBSYS_COUNT after reaching the end
588 * @ss_mask: the bitmask
590 * The block will only run for cases where the ssid-th bit (1 << ssid) of
593 #define do_each_subsys_mask(ss, ssid, ss_mask) do { \
594 unsigned long __ss_mask = (ss_mask); \
595 if (!CGROUP_SUBSYS_COUNT) { /* to avoid spurious gcc warning */ \
599 for_each_set_bit(ssid, &__ss_mask, CGROUP_SUBSYS_COUNT) { \
600 (ss) = cgroup_subsys[ssid]; \
603 #define while_each_subsys_mask() \
608 /* iterate over child cgrps, lock should be held throughout iteration */
609 #define cgroup_for_each_live_child(child, cgrp) \
610 list_for_each_entry((child), &(cgrp)->self.children, self.sibling) \
611 if (({ lockdep_assert_held(&cgroup_mutex); \
612 cgroup_is_dead(child); })) \
616 /* walk live descendants in preorder */
617 #define cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) \
618 css_for_each_descendant_pre((d_css), cgroup_css((cgrp), NULL)) \
619 if (({ lockdep_assert_held(&cgroup_mutex); \
620 (dsct) = (d_css)->cgroup; \
621 cgroup_is_dead(dsct); })) \
625 /* walk live descendants in postorder */
626 #define cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) \
627 css_for_each_descendant_post((d_css), cgroup_css((cgrp), NULL)) \
628 if (({ lockdep_assert_held(&cgroup_mutex); \
629 (dsct) = (d_css)->cgroup; \
630 cgroup_is_dead(dsct); })) \
635 * The default css_set - used by init and its children prior to any
636 * hierarchies being mounted. It contains a pointer to the root state
637 * for each subsystem. Also used to anchor the list of css_sets. Not
638 * reference-counted, to improve performance when child cgroups
639 * haven't been created.
641 struct css_set init_css_set = {
642 .refcount = REFCOUNT_INIT(1),
643 .dom_cset = &init_css_set,
644 .tasks = LIST_HEAD_INIT(init_css_set.tasks),
645 .mg_tasks = LIST_HEAD_INIT(init_css_set.mg_tasks),
646 .dying_tasks = LIST_HEAD_INIT(init_css_set.dying_tasks),
647 .task_iters = LIST_HEAD_INIT(init_css_set.task_iters),
648 .threaded_csets = LIST_HEAD_INIT(init_css_set.threaded_csets),
649 .cgrp_links = LIST_HEAD_INIT(init_css_set.cgrp_links),
650 .mg_preload_node = LIST_HEAD_INIT(init_css_set.mg_preload_node),
651 .mg_node = LIST_HEAD_INIT(init_css_set.mg_node),
654 static int css_set_count = 1; /* 1 for init_css_set */
656 static bool css_set_threaded(struct css_set *cset)
658 return cset->dom_cset != cset;
662 * css_set_populated - does a css_set contain any tasks?
663 * @cset: target css_set
665 * css_set_populated() should be the same as !!cset->nr_tasks at steady
666 * state. However, css_set_populated() can be called while a task is being
667 * added to or removed from the linked list before the nr_tasks is
668 * properly updated. Hence, we can't just look at ->nr_tasks here.
670 static bool css_set_populated(struct css_set *cset)
672 lockdep_assert_held(&css_set_lock);
674 return !list_empty(&cset->tasks) || !list_empty(&cset->mg_tasks);
678 * cgroup_update_populated - update the populated count of a cgroup
679 * @cgrp: the target cgroup
680 * @populated: inc or dec populated count
682 * One of the css_sets associated with @cgrp is either getting its first
683 * task or losing the last. Update @cgrp->nr_populated_* accordingly. The
684 * count is propagated towards root so that a given cgroup's
685 * nr_populated_children is zero iff none of its descendants contain any
688 * @cgrp's interface file "cgroup.populated" is zero if both
689 * @cgrp->nr_populated_csets and @cgrp->nr_populated_children are zero and
690 * 1 otherwise. When the sum changes from or to zero, userland is notified
691 * that the content of the interface file has changed. This can be used to
692 * detect when @cgrp and its descendants become populated or empty.
694 static void cgroup_update_populated(struct cgroup *cgrp, bool populated)
696 struct cgroup *child = NULL;
697 int adj = populated ? 1 : -1;
699 lockdep_assert_held(&css_set_lock);
702 bool was_populated = cgroup_is_populated(cgrp);
705 cgrp->nr_populated_csets += adj;
707 if (cgroup_is_threaded(child))
708 cgrp->nr_populated_threaded_children += adj;
710 cgrp->nr_populated_domain_children += adj;
713 if (was_populated == cgroup_is_populated(cgrp))
716 cgroup1_check_for_release(cgrp);
717 cgroup_file_notify(&cgrp->events_file);
720 cgrp = cgroup_parent(cgrp);
725 * css_set_update_populated - update populated state of a css_set
726 * @cset: target css_set
727 * @populated: whether @cset is populated or depopulated
729 * @cset is either getting the first task or losing the last. Update the
730 * populated counters of all associated cgroups accordingly.
732 static void css_set_update_populated(struct css_set *cset, bool populated)
734 struct cgrp_cset_link *link;
736 lockdep_assert_held(&css_set_lock);
738 list_for_each_entry(link, &cset->cgrp_links, cgrp_link)
739 cgroup_update_populated(link->cgrp, populated);
743 * @task is leaving, advance task iterators which are pointing to it so
744 * that they can resume at the next position. Advancing an iterator might
745 * remove it from the list, use safe walk. See css_task_iter_skip() for
748 static void css_set_skip_task_iters(struct css_set *cset,
749 struct task_struct *task)
751 struct css_task_iter *it, *pos;
753 list_for_each_entry_safe(it, pos, &cset->task_iters, iters_node)
754 css_task_iter_skip(it, task);
758 * css_set_move_task - move a task from one css_set to another
759 * @task: task being moved
760 * @from_cset: css_set @task currently belongs to (may be NULL)
761 * @to_cset: new css_set @task is being moved to (may be NULL)
762 * @use_mg_tasks: move to @to_cset->mg_tasks instead of ->tasks
764 * Move @task from @from_cset to @to_cset. If @task didn't belong to any
765 * css_set, @from_cset can be NULL. If @task is being disassociated
766 * instead of moved, @to_cset can be NULL.
768 * This function automatically handles populated counter updates and
769 * css_task_iter adjustments but the caller is responsible for managing
770 * @from_cset and @to_cset's reference counts.
772 static void css_set_move_task(struct task_struct *task,
773 struct css_set *from_cset, struct css_set *to_cset,
776 lockdep_assert_held(&css_set_lock);
778 if (to_cset && !css_set_populated(to_cset))
779 css_set_update_populated(to_cset, true);
782 WARN_ON_ONCE(list_empty(&task->cg_list));
784 css_set_skip_task_iters(from_cset, task);
785 list_del_init(&task->cg_list);
786 if (!css_set_populated(from_cset))
787 css_set_update_populated(from_cset, false);
789 WARN_ON_ONCE(!list_empty(&task->cg_list));
794 * We are synchronized through cgroup_threadgroup_rwsem
795 * against PF_EXITING setting such that we can't race
796 * against cgroup_exit() changing the css_set to
797 * init_css_set and dropping the old one.
799 WARN_ON_ONCE(task->flags & PF_EXITING);
801 rcu_assign_pointer(task->cgroups, to_cset);
802 list_add_tail(&task->cg_list, use_mg_tasks ? &to_cset->mg_tasks :
808 * hash table for cgroup groups. This improves the performance to find
809 * an existing css_set. This hash doesn't (currently) take into
810 * account cgroups in empty hierarchies.
812 #define CSS_SET_HASH_BITS 7
813 static DEFINE_HASHTABLE(css_set_table, CSS_SET_HASH_BITS);
815 static unsigned long css_set_hash(struct cgroup_subsys_state *css[])
817 unsigned long key = 0UL;
818 struct cgroup_subsys *ss;
821 for_each_subsys(ss, i)
822 key += (unsigned long)css[i];
823 key = (key >> 16) ^ key;
828 void put_css_set_locked(struct css_set *cset)
830 struct cgrp_cset_link *link, *tmp_link;
831 struct cgroup_subsys *ss;
834 lockdep_assert_held(&css_set_lock);
836 if (!refcount_dec_and_test(&cset->refcount))
839 WARN_ON_ONCE(!list_empty(&cset->threaded_csets));
841 /* This css_set is dead. unlink it and release cgroup and css refs */
842 for_each_subsys(ss, ssid) {
843 list_del(&cset->e_cset_node[ssid]);
844 css_put(cset->subsys[ssid]);
846 hash_del(&cset->hlist);
849 list_for_each_entry_safe(link, tmp_link, &cset->cgrp_links, cgrp_link) {
850 list_del(&link->cset_link);
851 list_del(&link->cgrp_link);
852 if (cgroup_parent(link->cgrp))
853 cgroup_put(link->cgrp);
857 if (css_set_threaded(cset)) {
858 list_del(&cset->threaded_csets_node);
859 put_css_set_locked(cset->dom_cset);
862 kfree_rcu(cset, rcu_head);
866 * compare_css_sets - helper function for find_existing_css_set().
867 * @cset: candidate css_set being tested
868 * @old_cset: existing css_set for a task
869 * @new_cgrp: cgroup that's being entered by the task
870 * @template: desired set of css pointers in css_set (pre-calculated)
872 * Returns true if "cset" matches "old_cset" except for the hierarchy
873 * which "new_cgrp" belongs to, for which it should match "new_cgrp".
875 static bool compare_css_sets(struct css_set *cset,
876 struct css_set *old_cset,
877 struct cgroup *new_cgrp,
878 struct cgroup_subsys_state *template[])
880 struct cgroup *new_dfl_cgrp;
881 struct list_head *l1, *l2;
884 * On the default hierarchy, there can be csets which are
885 * associated with the same set of cgroups but different csses.
886 * Let's first ensure that csses match.
888 if (memcmp(template, cset->subsys, sizeof(cset->subsys)))
892 /* @cset's domain should match the default cgroup's */
893 if (cgroup_on_dfl(new_cgrp))
894 new_dfl_cgrp = new_cgrp;
896 new_dfl_cgrp = old_cset->dfl_cgrp;
898 if (new_dfl_cgrp->dom_cgrp != cset->dom_cset->dfl_cgrp)
902 * Compare cgroup pointers in order to distinguish between
903 * different cgroups in hierarchies. As different cgroups may
904 * share the same effective css, this comparison is always
907 l1 = &cset->cgrp_links;
908 l2 = &old_cset->cgrp_links;
910 struct cgrp_cset_link *link1, *link2;
911 struct cgroup *cgrp1, *cgrp2;
915 /* See if we reached the end - both lists are equal length. */
916 if (l1 == &cset->cgrp_links) {
917 BUG_ON(l2 != &old_cset->cgrp_links);
920 BUG_ON(l2 == &old_cset->cgrp_links);
922 /* Locate the cgroups associated with these links. */
923 link1 = list_entry(l1, struct cgrp_cset_link, cgrp_link);
924 link2 = list_entry(l2, struct cgrp_cset_link, cgrp_link);
927 /* Hierarchies should be linked in the same order. */
928 BUG_ON(cgrp1->root != cgrp2->root);
931 * If this hierarchy is the hierarchy of the cgroup
932 * that's changing, then we need to check that this
933 * css_set points to the new cgroup; if it's any other
934 * hierarchy, then this css_set should point to the
935 * same cgroup as the old css_set.
937 if (cgrp1->root == new_cgrp->root) {
938 if (cgrp1 != new_cgrp)
949 * find_existing_css_set - init css array and find the matching css_set
950 * @old_cset: the css_set that we're using before the cgroup transition
951 * @cgrp: the cgroup that we're moving into
952 * @template: out param for the new set of csses, should be clear on entry
954 static struct css_set *find_existing_css_set(struct css_set *old_cset,
956 struct cgroup_subsys_state *template[])
958 struct cgroup_root *root = cgrp->root;
959 struct cgroup_subsys *ss;
960 struct css_set *cset;
965 * Build the set of subsystem state objects that we want to see in the
966 * new css_set. while subsystems can change globally, the entries here
967 * won't change, so no need for locking.
969 for_each_subsys(ss, i) {
970 if (root->subsys_mask & (1UL << i)) {
972 * @ss is in this hierarchy, so we want the
973 * effective css from @cgrp.
975 template[i] = cgroup_e_css(cgrp, ss);
978 * @ss is not in this hierarchy, so we don't want
981 template[i] = old_cset->subsys[i];
985 key = css_set_hash(template);
986 hash_for_each_possible(css_set_table, cset, hlist, key) {
987 if (!compare_css_sets(cset, old_cset, cgrp, template))
990 /* This css_set matches what we need */
994 /* No existing cgroup group matched */
998 static void free_cgrp_cset_links(struct list_head *links_to_free)
1000 struct cgrp_cset_link *link, *tmp_link;
1002 list_for_each_entry_safe(link, tmp_link, links_to_free, cset_link) {
1003 list_del(&link->cset_link);
1009 * allocate_cgrp_cset_links - allocate cgrp_cset_links
1010 * @count: the number of links to allocate
1011 * @tmp_links: list_head the allocated links are put on
1013 * Allocate @count cgrp_cset_link structures and chain them on @tmp_links
1014 * through ->cset_link. Returns 0 on success or -errno.
1016 static int allocate_cgrp_cset_links(int count, struct list_head *tmp_links)
1018 struct cgrp_cset_link *link;
1021 INIT_LIST_HEAD(tmp_links);
1023 for (i = 0; i < count; i++) {
1024 link = kzalloc(sizeof(*link), GFP_KERNEL);
1026 free_cgrp_cset_links(tmp_links);
1029 list_add(&link->cset_link, tmp_links);
1035 * link_css_set - a helper function to link a css_set to a cgroup
1036 * @tmp_links: cgrp_cset_link objects allocated by allocate_cgrp_cset_links()
1037 * @cset: the css_set to be linked
1038 * @cgrp: the destination cgroup
1040 static void link_css_set(struct list_head *tmp_links, struct css_set *cset,
1041 struct cgroup *cgrp)
1043 struct cgrp_cset_link *link;
1045 BUG_ON(list_empty(tmp_links));
1047 if (cgroup_on_dfl(cgrp))
1048 cset->dfl_cgrp = cgrp;
1050 link = list_first_entry(tmp_links, struct cgrp_cset_link, cset_link);
1055 * Always add links to the tail of the lists so that the lists are
1056 * in choronological order.
1058 list_move_tail(&link->cset_link, &cgrp->cset_links);
1059 list_add_tail(&link->cgrp_link, &cset->cgrp_links);
1061 if (cgroup_parent(cgrp))
1062 cgroup_get_live(cgrp);
1066 * find_css_set - return a new css_set with one cgroup updated
1067 * @old_cset: the baseline css_set
1068 * @cgrp: the cgroup to be updated
1070 * Return a new css_set that's equivalent to @old_cset, but with @cgrp
1071 * substituted into the appropriate hierarchy.
1073 static struct css_set *find_css_set(struct css_set *old_cset,
1074 struct cgroup *cgrp)
1076 struct cgroup_subsys_state *template[CGROUP_SUBSYS_COUNT] = { };
1077 struct css_set *cset;
1078 struct list_head tmp_links;
1079 struct cgrp_cset_link *link;
1080 struct cgroup_subsys *ss;
1084 lockdep_assert_held(&cgroup_mutex);
1086 /* First see if we already have a cgroup group that matches
1087 * the desired set */
1088 spin_lock_irq(&css_set_lock);
1089 cset = find_existing_css_set(old_cset, cgrp, template);
1092 spin_unlock_irq(&css_set_lock);
1097 cset = kzalloc(sizeof(*cset), GFP_KERNEL);
1101 /* Allocate all the cgrp_cset_link objects that we'll need */
1102 if (allocate_cgrp_cset_links(cgroup_root_count, &tmp_links) < 0) {
1107 refcount_set(&cset->refcount, 1);
1108 cset->dom_cset = cset;
1109 INIT_LIST_HEAD(&cset->tasks);
1110 INIT_LIST_HEAD(&cset->mg_tasks);
1111 INIT_LIST_HEAD(&cset->dying_tasks);
1112 INIT_LIST_HEAD(&cset->task_iters);
1113 INIT_LIST_HEAD(&cset->threaded_csets);
1114 INIT_HLIST_NODE(&cset->hlist);
1115 INIT_LIST_HEAD(&cset->cgrp_links);
1116 INIT_LIST_HEAD(&cset->mg_preload_node);
1117 INIT_LIST_HEAD(&cset->mg_node);
1119 /* Copy the set of subsystem state objects generated in
1120 * find_existing_css_set() */
1121 memcpy(cset->subsys, template, sizeof(cset->subsys));
1123 spin_lock_irq(&css_set_lock);
1124 /* Add reference counts and links from the new css_set. */
1125 list_for_each_entry(link, &old_cset->cgrp_links, cgrp_link) {
1126 struct cgroup *c = link->cgrp;
1128 if (c->root == cgrp->root)
1130 link_css_set(&tmp_links, cset, c);
1133 BUG_ON(!list_empty(&tmp_links));
1137 /* Add @cset to the hash table */
1138 key = css_set_hash(cset->subsys);
1139 hash_add(css_set_table, &cset->hlist, key);
1141 for_each_subsys(ss, ssid) {
1142 struct cgroup_subsys_state *css = cset->subsys[ssid];
1144 list_add_tail(&cset->e_cset_node[ssid],
1145 &css->cgroup->e_csets[ssid]);
1149 spin_unlock_irq(&css_set_lock);
1152 * If @cset should be threaded, look up the matching dom_cset and
1153 * link them up. We first fully initialize @cset then look for the
1154 * dom_cset. It's simpler this way and safe as @cset is guaranteed
1155 * to stay empty until we return.
1157 if (cgroup_is_threaded(cset->dfl_cgrp)) {
1158 struct css_set *dcset;
1160 dcset = find_css_set(cset, cset->dfl_cgrp->dom_cgrp);
1166 spin_lock_irq(&css_set_lock);
1167 cset->dom_cset = dcset;
1168 list_add_tail(&cset->threaded_csets_node,
1169 &dcset->threaded_csets);
1170 spin_unlock_irq(&css_set_lock);
1176 struct cgroup_root *cgroup_root_from_kf(struct kernfs_root *kf_root)
1178 struct cgroup *root_cgrp = kf_root->kn->priv;
1180 return root_cgrp->root;
1183 static int cgroup_init_root_id(struct cgroup_root *root)
1187 lockdep_assert_held(&cgroup_mutex);
1189 id = idr_alloc_cyclic(&cgroup_hierarchy_idr, root, 0, 0, GFP_KERNEL);
1193 root->hierarchy_id = id;
1197 static void cgroup_exit_root_id(struct cgroup_root *root)
1199 lockdep_assert_held(&cgroup_mutex);
1201 idr_remove(&cgroup_hierarchy_idr, root->hierarchy_id);
1204 void cgroup_free_root(struct cgroup_root *root)
1207 idr_destroy(&root->cgroup_idr);
1212 static void cgroup_destroy_root(struct cgroup_root *root)
1214 struct cgroup *cgrp = &root->cgrp;
1215 struct cgrp_cset_link *link, *tmp_link;
1217 trace_cgroup_destroy_root(root);
1219 cgroup_lock_and_drain_offline(&cgrp_dfl_root.cgrp);
1221 BUG_ON(atomic_read(&root->nr_cgrps));
1222 BUG_ON(!list_empty(&cgrp->self.children));
1224 /* Rebind all subsystems back to the default hierarchy */
1225 WARN_ON(rebind_subsystems(&cgrp_dfl_root, root->subsys_mask));
1228 * Release all the links from cset_links to this hierarchy's
1231 spin_lock_irq(&css_set_lock);
1233 list_for_each_entry_safe(link, tmp_link, &cgrp->cset_links, cset_link) {
1234 list_del(&link->cset_link);
1235 list_del(&link->cgrp_link);
1239 spin_unlock_irq(&css_set_lock);
1241 if (!list_empty(&root->root_list)) {
1242 list_del(&root->root_list);
1243 cgroup_root_count--;
1246 cgroup_exit_root_id(root);
1248 mutex_unlock(&cgroup_mutex);
1250 kernfs_destroy_root(root->kf_root);
1251 cgroup_free_root(root);
1255 * look up cgroup associated with current task's cgroup namespace on the
1256 * specified hierarchy
1258 static struct cgroup *
1259 current_cgns_cgroup_from_root(struct cgroup_root *root)
1261 struct cgroup *res = NULL;
1262 struct css_set *cset;
1264 lockdep_assert_held(&css_set_lock);
1268 cset = current->nsproxy->cgroup_ns->root_cset;
1269 if (cset == &init_css_set) {
1272 struct cgrp_cset_link *link;
1274 list_for_each_entry(link, &cset->cgrp_links, cgrp_link) {
1275 struct cgroup *c = link->cgrp;
1277 if (c->root == root) {
1289 /* look up cgroup associated with given css_set on the specified hierarchy */
1290 static struct cgroup *cset_cgroup_from_root(struct css_set *cset,
1291 struct cgroup_root *root)
1293 struct cgroup *res = NULL;
1295 lockdep_assert_held(&cgroup_mutex);
1296 lockdep_assert_held(&css_set_lock);
1298 if (cset == &init_css_set) {
1300 } else if (root == &cgrp_dfl_root) {
1301 res = cset->dfl_cgrp;
1303 struct cgrp_cset_link *link;
1305 list_for_each_entry(link, &cset->cgrp_links, cgrp_link) {
1306 struct cgroup *c = link->cgrp;
1308 if (c->root == root) {
1320 * Return the cgroup for "task" from the given hierarchy. Must be
1321 * called with cgroup_mutex and css_set_lock held.
1323 struct cgroup *task_cgroup_from_root(struct task_struct *task,
1324 struct cgroup_root *root)
1327 * No need to lock the task - since we hold cgroup_mutex the
1328 * task can't change groups, so the only thing that can happen
1329 * is that it exits and its css is set back to init_css_set.
1331 return cset_cgroup_from_root(task_css_set(task), root);
1335 * A task must hold cgroup_mutex to modify cgroups.
1337 * Any task can increment and decrement the count field without lock.
1338 * So in general, code holding cgroup_mutex can't rely on the count
1339 * field not changing. However, if the count goes to zero, then only
1340 * cgroup_attach_task() can increment it again. Because a count of zero
1341 * means that no tasks are currently attached, therefore there is no
1342 * way a task attached to that cgroup can fork (the other way to
1343 * increment the count). So code holding cgroup_mutex can safely
1344 * assume that if the count is zero, it will stay zero. Similarly, if
1345 * a task holds cgroup_mutex on a cgroup with zero count, it
1346 * knows that the cgroup won't be removed, as cgroup_rmdir()
1349 * A cgroup can only be deleted if both its 'count' of using tasks
1350 * is zero, and its list of 'children' cgroups is empty. Since all
1351 * tasks in the system use _some_ cgroup, and since there is always at
1352 * least one task in the system (init, pid == 1), therefore, root cgroup
1353 * always has either children cgroups and/or using tasks. So we don't
1354 * need a special hack to ensure that root cgroup cannot be deleted.
1356 * P.S. One more locking exception. RCU is used to guard the
1357 * update of a tasks cgroup pointer by cgroup_attach_task()
1360 static struct kernfs_syscall_ops cgroup_kf_syscall_ops;
1362 static char *cgroup_file_name(struct cgroup *cgrp, const struct cftype *cft,
1365 struct cgroup_subsys *ss = cft->ss;
1367 if (cft->ss && !(cft->flags & CFTYPE_NO_PREFIX) &&
1368 !(cgrp->root->flags & CGRP_ROOT_NOPREFIX))
1369 snprintf(buf, CGROUP_FILE_NAME_MAX, "%s.%s",
1370 cgroup_on_dfl(cgrp) ? ss->name : ss->legacy_name,
1373 strncpy(buf, cft->name, CGROUP_FILE_NAME_MAX);
1378 * cgroup_file_mode - deduce file mode of a control file
1379 * @cft: the control file in question
1381 * S_IRUGO for read, S_IWUSR for write.
1383 static umode_t cgroup_file_mode(const struct cftype *cft)
1387 if (cft->read_u64 || cft->read_s64 || cft->seq_show)
1390 if (cft->write_u64 || cft->write_s64 || cft->write) {
1391 if (cft->flags & CFTYPE_WORLD_WRITABLE)
1401 * cgroup_calc_subtree_ss_mask - calculate subtree_ss_mask
1402 * @subtree_control: the new subtree_control mask to consider
1403 * @this_ss_mask: available subsystems
1405 * On the default hierarchy, a subsystem may request other subsystems to be
1406 * enabled together through its ->depends_on mask. In such cases, more
1407 * subsystems than specified in "cgroup.subtree_control" may be enabled.
1409 * This function calculates which subsystems need to be enabled if
1410 * @subtree_control is to be applied while restricted to @this_ss_mask.
1412 static u16 cgroup_calc_subtree_ss_mask(u16 subtree_control, u16 this_ss_mask)
1414 u16 cur_ss_mask = subtree_control;
1415 struct cgroup_subsys *ss;
1418 lockdep_assert_held(&cgroup_mutex);
1420 cur_ss_mask |= cgrp_dfl_implicit_ss_mask;
1423 u16 new_ss_mask = cur_ss_mask;
1425 do_each_subsys_mask(ss, ssid, cur_ss_mask) {
1426 new_ss_mask |= ss->depends_on;
1427 } while_each_subsys_mask();
1430 * Mask out subsystems which aren't available. This can
1431 * happen only if some depended-upon subsystems were bound
1432 * to non-default hierarchies.
1434 new_ss_mask &= this_ss_mask;
1436 if (new_ss_mask == cur_ss_mask)
1438 cur_ss_mask = new_ss_mask;
1445 * cgroup_kn_unlock - unlocking helper for cgroup kernfs methods
1446 * @kn: the kernfs_node being serviced
1448 * This helper undoes cgroup_kn_lock_live() and should be invoked before
1449 * the method finishes if locking succeeded. Note that once this function
1450 * returns the cgroup returned by cgroup_kn_lock_live() may become
1451 * inaccessible any time. If the caller intends to continue to access the
1452 * cgroup, it should pin it before invoking this function.
1454 void cgroup_kn_unlock(struct kernfs_node *kn)
1456 struct cgroup *cgrp;
1458 if (kernfs_type(kn) == KERNFS_DIR)
1461 cgrp = kn->parent->priv;
1463 mutex_unlock(&cgroup_mutex);
1465 kernfs_unbreak_active_protection(kn);
1470 * cgroup_kn_lock_live - locking helper for cgroup kernfs methods
1471 * @kn: the kernfs_node being serviced
1472 * @drain_offline: perform offline draining on the cgroup
1474 * This helper is to be used by a cgroup kernfs method currently servicing
1475 * @kn. It breaks the active protection, performs cgroup locking and
1476 * verifies that the associated cgroup is alive. Returns the cgroup if
1477 * alive; otherwise, %NULL. A successful return should be undone by a
1478 * matching cgroup_kn_unlock() invocation. If @drain_offline is %true, the
1479 * cgroup is drained of offlining csses before return.
1481 * Any cgroup kernfs method implementation which requires locking the
1482 * associated cgroup should use this helper. It avoids nesting cgroup
1483 * locking under kernfs active protection and allows all kernfs operations
1484 * including self-removal.
1486 struct cgroup *cgroup_kn_lock_live(struct kernfs_node *kn, bool drain_offline)
1488 struct cgroup *cgrp;
1490 if (kernfs_type(kn) == KERNFS_DIR)
1493 cgrp = kn->parent->priv;
1496 * We're gonna grab cgroup_mutex which nests outside kernfs
1497 * active_ref. cgroup liveliness check alone provides enough
1498 * protection against removal. Ensure @cgrp stays accessible and
1499 * break the active_ref protection.
1501 if (!cgroup_tryget(cgrp))
1503 kernfs_break_active_protection(kn);
1506 cgroup_lock_and_drain_offline(cgrp);
1508 mutex_lock(&cgroup_mutex);
1510 if (!cgroup_is_dead(cgrp))
1513 cgroup_kn_unlock(kn);
1517 static void cgroup_rm_file(struct cgroup *cgrp, const struct cftype *cft)
1519 char name[CGROUP_FILE_NAME_MAX];
1521 lockdep_assert_held(&cgroup_mutex);
1523 if (cft->file_offset) {
1524 struct cgroup_subsys_state *css = cgroup_css(cgrp, cft->ss);
1525 struct cgroup_file *cfile = (void *)css + cft->file_offset;
1527 spin_lock_irq(&cgroup_file_kn_lock);
1529 spin_unlock_irq(&cgroup_file_kn_lock);
1532 kernfs_remove_by_name(cgrp->kn, cgroup_file_name(cgrp, cft, name));
1536 * css_clear_dir - remove subsys files in a cgroup directory
1539 static void css_clear_dir(struct cgroup_subsys_state *css)
1541 struct cgroup *cgrp = css->cgroup;
1542 struct cftype *cfts;
1544 if (!(css->flags & CSS_VISIBLE))
1547 css->flags &= ~CSS_VISIBLE;
1549 list_for_each_entry(cfts, &css->ss->cfts, node)
1550 cgroup_addrm_files(css, cgrp, cfts, false);
1554 * css_populate_dir - create subsys files in a cgroup directory
1557 * On failure, no file is added.
1559 static int css_populate_dir(struct cgroup_subsys_state *css)
1561 struct cgroup *cgrp = css->cgroup;
1562 struct cftype *cfts, *failed_cfts;
1565 if ((css->flags & CSS_VISIBLE) || !cgrp->kn)
1569 if (cgroup_on_dfl(cgrp))
1570 cfts = cgroup_base_files;
1572 cfts = cgroup1_base_files;
1574 return cgroup_addrm_files(&cgrp->self, cgrp, cfts, true);
1577 list_for_each_entry(cfts, &css->ss->cfts, node) {
1578 ret = cgroup_addrm_files(css, cgrp, cfts, true);
1585 css->flags |= CSS_VISIBLE;
1589 list_for_each_entry(cfts, &css->ss->cfts, node) {
1590 if (cfts == failed_cfts)
1592 cgroup_addrm_files(css, cgrp, cfts, false);
1597 int rebind_subsystems(struct cgroup_root *dst_root, u16 ss_mask)
1599 struct cgroup *dcgrp = &dst_root->cgrp;
1600 struct cgroup_subsys *ss;
1602 u16 dfl_disable_ss_mask = 0;
1604 lockdep_assert_held(&cgroup_mutex);
1606 do_each_subsys_mask(ss, ssid, ss_mask) {
1608 * If @ss has non-root csses attached to it, can't move.
1609 * If @ss is an implicit controller, it is exempt from this
1610 * rule and can be stolen.
1612 if (css_next_child(NULL, cgroup_css(&ss->root->cgrp, ss)) &&
1613 !ss->implicit_on_dfl)
1616 /* can't move between two non-dummy roots either */
1617 if (ss->root != &cgrp_dfl_root && dst_root != &cgrp_dfl_root)
1621 * Collect ssid's that need to be disabled from default
1624 if (ss->root == &cgrp_dfl_root)
1625 dfl_disable_ss_mask |= 1 << ssid;
1627 } while_each_subsys_mask();
1629 if (dfl_disable_ss_mask) {
1630 struct cgroup *scgrp = &cgrp_dfl_root.cgrp;
1633 * Controllers from default hierarchy that need to be rebound
1634 * are all disabled together in one go.
1636 cgrp_dfl_root.subsys_mask &= ~dfl_disable_ss_mask;
1637 WARN_ON(cgroup_apply_control(scgrp));
1638 cgroup_finalize_control(scgrp, 0);
1641 do_each_subsys_mask(ss, ssid, ss_mask) {
1642 struct cgroup_root *src_root = ss->root;
1643 struct cgroup *scgrp = &src_root->cgrp;
1644 struct cgroup_subsys_state *css = cgroup_css(scgrp, ss);
1645 struct css_set *cset;
1647 WARN_ON(!css || cgroup_css(dcgrp, ss));
1649 if (src_root != &cgrp_dfl_root) {
1650 /* disable from the source */
1651 src_root->subsys_mask &= ~(1 << ssid);
1652 WARN_ON(cgroup_apply_control(scgrp));
1653 cgroup_finalize_control(scgrp, 0);
1657 RCU_INIT_POINTER(scgrp->subsys[ssid], NULL);
1658 rcu_assign_pointer(dcgrp->subsys[ssid], css);
1659 ss->root = dst_root;
1660 css->cgroup = dcgrp;
1662 spin_lock_irq(&css_set_lock);
1663 hash_for_each(css_set_table, i, cset, hlist)
1664 list_move_tail(&cset->e_cset_node[ss->id],
1665 &dcgrp->e_csets[ss->id]);
1666 spin_unlock_irq(&css_set_lock);
1668 /* default hierarchy doesn't enable controllers by default */
1669 dst_root->subsys_mask |= 1 << ssid;
1670 if (dst_root == &cgrp_dfl_root) {
1671 static_branch_enable(cgroup_subsys_on_dfl_key[ssid]);
1673 dcgrp->subtree_control |= 1 << ssid;
1674 static_branch_disable(cgroup_subsys_on_dfl_key[ssid]);
1677 ret = cgroup_apply_control(dcgrp);
1679 pr_warn("partial failure to rebind %s controller (err=%d)\n",
1684 } while_each_subsys_mask();
1686 kernfs_activate(dcgrp->kn);
1690 int cgroup_show_path(struct seq_file *sf, struct kernfs_node *kf_node,
1691 struct kernfs_root *kf_root)
1695 struct cgroup_root *kf_cgroot = cgroup_root_from_kf(kf_root);
1696 struct cgroup *ns_cgroup;
1698 buf = kmalloc(PATH_MAX, GFP_KERNEL);
1702 spin_lock_irq(&css_set_lock);
1703 ns_cgroup = current_cgns_cgroup_from_root(kf_cgroot);
1704 len = kernfs_path_from_node(kf_node, ns_cgroup->kn, buf, PATH_MAX);
1705 spin_unlock_irq(&css_set_lock);
1707 if (len >= PATH_MAX)
1710 seq_escape(sf, buf, " \t\n\\");
1717 static int parse_cgroup_root_flags(char *data, unsigned int *root_flags)
1723 if (!data || *data == '\0')
1726 while ((token = strsep(&data, ",")) != NULL) {
1727 if (!strcmp(token, "nsdelegate")) {
1728 *root_flags |= CGRP_ROOT_NS_DELEGATE;
1732 pr_err("cgroup2: unknown option \"%s\"\n", token);
1739 static void apply_cgroup_root_flags(unsigned int root_flags)
1741 if (current->nsproxy->cgroup_ns == &init_cgroup_ns) {
1742 if (root_flags & CGRP_ROOT_NS_DELEGATE)
1743 cgrp_dfl_root.flags |= CGRP_ROOT_NS_DELEGATE;
1745 cgrp_dfl_root.flags &= ~CGRP_ROOT_NS_DELEGATE;
1749 static int cgroup_show_options(struct seq_file *seq, struct kernfs_root *kf_root)
1751 if (cgrp_dfl_root.flags & CGRP_ROOT_NS_DELEGATE)
1752 seq_puts(seq, ",nsdelegate");
1756 static int cgroup_remount(struct kernfs_root *kf_root, int *flags, char *data)
1758 unsigned int root_flags;
1761 ret = parse_cgroup_root_flags(data, &root_flags);
1765 apply_cgroup_root_flags(root_flags);
1770 * To reduce the fork() overhead for systems that are not actually using
1771 * their cgroups capability, we don't maintain the lists running through
1772 * each css_set to its tasks until we see the list actually used - in other
1773 * words after the first mount.
1775 static bool use_task_css_set_links __read_mostly;
1777 static void cgroup_enable_task_cg_lists(void)
1779 struct task_struct *p, *g;
1781 spin_lock_irq(&css_set_lock);
1783 if (use_task_css_set_links)
1786 use_task_css_set_links = true;
1789 * We need tasklist_lock because RCU is not safe against
1790 * while_each_thread(). Besides, a forking task that has passed
1791 * cgroup_post_fork() without seeing use_task_css_set_links = 1
1792 * is not guaranteed to have its child immediately visible in the
1793 * tasklist if we walk through it with RCU.
1795 read_lock(&tasklist_lock);
1796 do_each_thread(g, p) {
1797 WARN_ON_ONCE(!list_empty(&p->cg_list) ||
1798 task_css_set(p) != &init_css_set);
1801 * We should check if the process is exiting, otherwise
1802 * it will race with cgroup_exit() in that the list
1803 * entry won't be deleted though the process has exited.
1804 * Do it while holding siglock so that we don't end up
1805 * racing against cgroup_exit().
1807 * Interrupts were already disabled while acquiring
1808 * the css_set_lock, so we do not need to disable it
1809 * again when acquiring the sighand->siglock here.
1811 spin_lock(&p->sighand->siglock);
1812 if (!(p->flags & PF_EXITING)) {
1813 struct css_set *cset = task_css_set(p);
1815 if (!css_set_populated(cset))
1816 css_set_update_populated(cset, true);
1817 list_add_tail(&p->cg_list, &cset->tasks);
1821 spin_unlock(&p->sighand->siglock);
1822 } while_each_thread(g, p);
1823 read_unlock(&tasklist_lock);
1825 spin_unlock_irq(&css_set_lock);
1828 static void init_cgroup_housekeeping(struct cgroup *cgrp)
1830 struct cgroup_subsys *ss;
1833 INIT_LIST_HEAD(&cgrp->self.sibling);
1834 INIT_LIST_HEAD(&cgrp->self.children);
1835 INIT_LIST_HEAD(&cgrp->cset_links);
1836 INIT_LIST_HEAD(&cgrp->pidlists);
1837 mutex_init(&cgrp->pidlist_mutex);
1838 cgrp->self.cgroup = cgrp;
1839 cgrp->self.flags |= CSS_ONLINE;
1840 cgrp->dom_cgrp = cgrp;
1841 cgrp->max_descendants = INT_MAX;
1842 cgrp->max_depth = INT_MAX;
1844 for_each_subsys(ss, ssid)
1845 INIT_LIST_HEAD(&cgrp->e_csets[ssid]);
1847 init_waitqueue_head(&cgrp->offline_waitq);
1848 INIT_WORK(&cgrp->release_agent_work, cgroup1_release_agent);
1851 void init_cgroup_root(struct cgroup_root *root, struct cgroup_sb_opts *opts)
1853 struct cgroup *cgrp = &root->cgrp;
1855 INIT_LIST_HEAD(&root->root_list);
1856 atomic_set(&root->nr_cgrps, 1);
1858 init_cgroup_housekeeping(cgrp);
1859 idr_init(&root->cgroup_idr);
1861 root->flags = opts->flags;
1862 if (opts->release_agent)
1863 strcpy(root->release_agent_path, opts->release_agent);
1865 strcpy(root->name, opts->name);
1866 if (opts->cpuset_clone_children)
1867 set_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->cgrp.flags);
1870 int cgroup_setup_root(struct cgroup_root *root, u16 ss_mask, int ref_flags)
1872 LIST_HEAD(tmp_links);
1873 struct cgroup *root_cgrp = &root->cgrp;
1874 struct kernfs_syscall_ops *kf_sops;
1875 struct css_set *cset;
1878 lockdep_assert_held(&cgroup_mutex);
1880 ret = cgroup_idr_alloc(&root->cgroup_idr, root_cgrp, 1, 2, GFP_KERNEL);
1883 root_cgrp->id = ret;
1884 root_cgrp->ancestor_ids[0] = ret;
1886 ret = percpu_ref_init(&root_cgrp->self.refcnt, css_release,
1887 ref_flags, GFP_KERNEL);
1892 * We're accessing css_set_count without locking css_set_lock here,
1893 * but that's OK - it can only be increased by someone holding
1894 * cgroup_lock, and that's us. Later rebinding may disable
1895 * controllers on the default hierarchy and thus create new csets,
1896 * which can't be more than the existing ones. Allocate 2x.
1898 ret = allocate_cgrp_cset_links(2 * css_set_count, &tmp_links);
1902 ret = cgroup_init_root_id(root);
1906 kf_sops = root == &cgrp_dfl_root ?
1907 &cgroup_kf_syscall_ops : &cgroup1_kf_syscall_ops;
1909 root->kf_root = kernfs_create_root(kf_sops,
1910 KERNFS_ROOT_CREATE_DEACTIVATED |
1911 KERNFS_ROOT_SUPPORT_EXPORTOP,
1913 if (IS_ERR(root->kf_root)) {
1914 ret = PTR_ERR(root->kf_root);
1917 root_cgrp->kn = root->kf_root->kn;
1919 ret = css_populate_dir(&root_cgrp->self);
1923 ret = rebind_subsystems(root, ss_mask);
1927 trace_cgroup_setup_root(root);
1930 * There must be no failure case after here, since rebinding takes
1931 * care of subsystems' refcounts, which are explicitly dropped in
1932 * the failure exit path.
1934 list_add(&root->root_list, &cgroup_roots);
1935 cgroup_root_count++;
1938 * Link the root cgroup in this hierarchy into all the css_set
1941 spin_lock_irq(&css_set_lock);
1942 hash_for_each(css_set_table, i, cset, hlist) {
1943 link_css_set(&tmp_links, cset, root_cgrp);
1944 if (css_set_populated(cset))
1945 cgroup_update_populated(root_cgrp, true);
1947 spin_unlock_irq(&css_set_lock);
1949 BUG_ON(!list_empty(&root_cgrp->self.children));
1950 BUG_ON(atomic_read(&root->nr_cgrps) != 1);
1952 kernfs_activate(root_cgrp->kn);
1957 kernfs_destroy_root(root->kf_root);
1958 root->kf_root = NULL;
1960 cgroup_exit_root_id(root);
1962 percpu_ref_exit(&root_cgrp->self.refcnt);
1964 free_cgrp_cset_links(&tmp_links);
1968 struct dentry *cgroup_do_mount(struct file_system_type *fs_type, int flags,
1969 struct cgroup_root *root, unsigned long magic,
1970 struct cgroup_namespace *ns)
1972 struct dentry *dentry;
1973 bool new_sb = false;
1975 dentry = kernfs_mount(fs_type, flags, root->kf_root, magic, &new_sb);
1978 * In non-init cgroup namespace, instead of root cgroup's dentry,
1979 * we return the dentry corresponding to the cgroupns->root_cgrp.
1981 if (!IS_ERR(dentry) && ns != &init_cgroup_ns) {
1982 struct dentry *nsdentry;
1983 struct super_block *sb = dentry->d_sb;
1984 struct cgroup *cgrp;
1986 mutex_lock(&cgroup_mutex);
1987 spin_lock_irq(&css_set_lock);
1989 cgrp = cset_cgroup_from_root(ns->root_cset, root);
1991 spin_unlock_irq(&css_set_lock);
1992 mutex_unlock(&cgroup_mutex);
1994 nsdentry = kernfs_node_dentry(cgrp->kn, sb);
1996 if (IS_ERR(nsdentry))
1997 deactivate_locked_super(sb);
2002 cgroup_put(&root->cgrp);
2007 static struct dentry *cgroup_mount(struct file_system_type *fs_type,
2008 int flags, const char *unused_dev_name,
2011 struct cgroup_namespace *ns = current->nsproxy->cgroup_ns;
2012 struct dentry *dentry;
2017 /* Check if the caller has permission to mount. */
2018 if (!ns_capable(ns->user_ns, CAP_SYS_ADMIN)) {
2020 return ERR_PTR(-EPERM);
2024 * The first time anyone tries to mount a cgroup, enable the list
2025 * linking each css_set to its tasks and fix up all existing tasks.
2027 if (!use_task_css_set_links)
2028 cgroup_enable_task_cg_lists();
2030 if (fs_type == &cgroup2_fs_type) {
2031 unsigned int root_flags;
2033 ret = parse_cgroup_root_flags(data, &root_flags);
2036 return ERR_PTR(ret);
2039 cgrp_dfl_visible = true;
2040 cgroup_get_live(&cgrp_dfl_root.cgrp);
2042 dentry = cgroup_do_mount(&cgroup2_fs_type, flags, &cgrp_dfl_root,
2043 CGROUP2_SUPER_MAGIC, ns);
2044 if (!IS_ERR(dentry))
2045 apply_cgroup_root_flags(root_flags);
2047 dentry = cgroup1_mount(&cgroup_fs_type, flags, data,
2048 CGROUP_SUPER_MAGIC, ns);
2055 static void cgroup_kill_sb(struct super_block *sb)
2057 struct kernfs_root *kf_root = kernfs_root_from_sb(sb);
2058 struct cgroup_root *root = cgroup_root_from_kf(kf_root);
2061 * If @root doesn't have any mounts or children, start killing it.
2062 * This prevents new mounts by disabling percpu_ref_tryget_live().
2063 * cgroup_mount() may wait for @root's release.
2065 * And don't kill the default root.
2067 if (!list_empty(&root->cgrp.self.children) ||
2068 root == &cgrp_dfl_root)
2069 cgroup_put(&root->cgrp);
2071 percpu_ref_kill(&root->cgrp.self.refcnt);
2076 struct file_system_type cgroup_fs_type = {
2078 .mount = cgroup_mount,
2079 .kill_sb = cgroup_kill_sb,
2080 .fs_flags = FS_USERNS_MOUNT,
2083 static struct file_system_type cgroup2_fs_type = {
2085 .mount = cgroup_mount,
2086 .kill_sb = cgroup_kill_sb,
2087 .fs_flags = FS_USERNS_MOUNT,
2090 int cgroup_path_ns_locked(struct cgroup *cgrp, char *buf, size_t buflen,
2091 struct cgroup_namespace *ns)
2093 struct cgroup *root = cset_cgroup_from_root(ns->root_cset, cgrp->root);
2095 return kernfs_path_from_node(cgrp->kn, root->kn, buf, buflen);
2098 int cgroup_path_ns(struct cgroup *cgrp, char *buf, size_t buflen,
2099 struct cgroup_namespace *ns)
2103 mutex_lock(&cgroup_mutex);
2104 spin_lock_irq(&css_set_lock);
2106 ret = cgroup_path_ns_locked(cgrp, buf, buflen, ns);
2108 spin_unlock_irq(&css_set_lock);
2109 mutex_unlock(&cgroup_mutex);
2113 EXPORT_SYMBOL_GPL(cgroup_path_ns);
2116 * task_cgroup_path - cgroup path of a task in the first cgroup hierarchy
2117 * @task: target task
2118 * @buf: the buffer to write the path into
2119 * @buflen: the length of the buffer
2121 * Determine @task's cgroup on the first (the one with the lowest non-zero
2122 * hierarchy_id) cgroup hierarchy and copy its path into @buf. This
2123 * function grabs cgroup_mutex and shouldn't be used inside locks used by
2124 * cgroup controller callbacks.
2126 * Return value is the same as kernfs_path().
2128 int task_cgroup_path(struct task_struct *task, char *buf, size_t buflen)
2130 struct cgroup_root *root;
2131 struct cgroup *cgrp;
2132 int hierarchy_id = 1;
2135 mutex_lock(&cgroup_mutex);
2136 spin_lock_irq(&css_set_lock);
2138 root = idr_get_next(&cgroup_hierarchy_idr, &hierarchy_id);
2141 cgrp = task_cgroup_from_root(task, root);
2142 ret = cgroup_path_ns_locked(cgrp, buf, buflen, &init_cgroup_ns);
2144 /* if no hierarchy exists, everyone is in "/" */
2145 ret = strlcpy(buf, "/", buflen);
2148 spin_unlock_irq(&css_set_lock);
2149 mutex_unlock(&cgroup_mutex);
2152 EXPORT_SYMBOL_GPL(task_cgroup_path);
2155 * cgroup_migrate_add_task - add a migration target task to a migration context
2156 * @task: target task
2157 * @mgctx: target migration context
2159 * Add @task, which is a migration target, to @mgctx->tset. This function
2160 * becomes noop if @task doesn't need to be migrated. @task's css_set
2161 * should have been added as a migration source and @task->cg_list will be
2162 * moved from the css_set's tasks list to mg_tasks one.
2164 static void cgroup_migrate_add_task(struct task_struct *task,
2165 struct cgroup_mgctx *mgctx)
2167 struct css_set *cset;
2169 lockdep_assert_held(&css_set_lock);
2171 /* @task either already exited or can't exit until the end */
2172 if (task->flags & PF_EXITING)
2175 /* leave @task alone if post_fork() hasn't linked it yet */
2176 if (list_empty(&task->cg_list))
2179 cset = task_css_set(task);
2180 if (!cset->mg_src_cgrp)
2183 mgctx->tset.nr_tasks++;
2185 list_move_tail(&task->cg_list, &cset->mg_tasks);
2186 if (list_empty(&cset->mg_node))
2187 list_add_tail(&cset->mg_node,
2188 &mgctx->tset.src_csets);
2189 if (list_empty(&cset->mg_dst_cset->mg_node))
2190 list_add_tail(&cset->mg_dst_cset->mg_node,
2191 &mgctx->tset.dst_csets);
2195 * cgroup_taskset_first - reset taskset and return the first task
2196 * @tset: taskset of interest
2197 * @dst_cssp: output variable for the destination css
2199 * @tset iteration is initialized and the first task is returned.
2201 struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset,
2202 struct cgroup_subsys_state **dst_cssp)
2204 tset->cur_cset = list_first_entry(tset->csets, struct css_set, mg_node);
2205 tset->cur_task = NULL;
2207 return cgroup_taskset_next(tset, dst_cssp);
2211 * cgroup_taskset_next - iterate to the next task in taskset
2212 * @tset: taskset of interest
2213 * @dst_cssp: output variable for the destination css
2215 * Return the next task in @tset. Iteration must have been initialized
2216 * with cgroup_taskset_first().
2218 struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset,
2219 struct cgroup_subsys_state **dst_cssp)
2221 struct css_set *cset = tset->cur_cset;
2222 struct task_struct *task = tset->cur_task;
2224 while (&cset->mg_node != tset->csets) {
2226 task = list_first_entry(&cset->mg_tasks,
2227 struct task_struct, cg_list);
2229 task = list_next_entry(task, cg_list);
2231 if (&task->cg_list != &cset->mg_tasks) {
2232 tset->cur_cset = cset;
2233 tset->cur_task = task;
2236 * This function may be called both before and
2237 * after cgroup_taskset_migrate(). The two cases
2238 * can be distinguished by looking at whether @cset
2239 * has its ->mg_dst_cset set.
2241 if (cset->mg_dst_cset)
2242 *dst_cssp = cset->mg_dst_cset->subsys[tset->ssid];
2244 *dst_cssp = cset->subsys[tset->ssid];
2249 cset = list_next_entry(cset, mg_node);
2257 * cgroup_taskset_migrate - migrate a taskset
2258 * @mgctx: migration context
2260 * Migrate tasks in @mgctx as setup by migration preparation functions.
2261 * This function fails iff one of the ->can_attach callbacks fails and
2262 * guarantees that either all or none of the tasks in @mgctx are migrated.
2263 * @mgctx is consumed regardless of success.
2265 static int cgroup_migrate_execute(struct cgroup_mgctx *mgctx)
2267 struct cgroup_taskset *tset = &mgctx->tset;
2268 struct cgroup_subsys *ss;
2269 struct task_struct *task, *tmp_task;
2270 struct css_set *cset, *tmp_cset;
2271 int ssid, failed_ssid, ret;
2273 /* check that we can legitimately attach to the cgroup */
2274 if (tset->nr_tasks) {
2275 do_each_subsys_mask(ss, ssid, mgctx->ss_mask) {
2276 if (ss->can_attach) {
2278 ret = ss->can_attach(tset);
2281 goto out_cancel_attach;
2284 } while_each_subsys_mask();
2288 * Now that we're guaranteed success, proceed to move all tasks to
2289 * the new cgroup. There are no failure cases after here, so this
2290 * is the commit point.
2292 spin_lock_irq(&css_set_lock);
2293 list_for_each_entry(cset, &tset->src_csets, mg_node) {
2294 list_for_each_entry_safe(task, tmp_task, &cset->mg_tasks, cg_list) {
2295 struct css_set *from_cset = task_css_set(task);
2296 struct css_set *to_cset = cset->mg_dst_cset;
2298 get_css_set(to_cset);
2299 to_cset->nr_tasks++;
2300 css_set_move_task(task, from_cset, to_cset, true);
2301 put_css_set_locked(from_cset);
2302 from_cset->nr_tasks--;
2305 spin_unlock_irq(&css_set_lock);
2308 * Migration is committed, all target tasks are now on dst_csets.
2309 * Nothing is sensitive to fork() after this point. Notify
2310 * controllers that migration is complete.
2312 tset->csets = &tset->dst_csets;
2314 if (tset->nr_tasks) {
2315 do_each_subsys_mask(ss, ssid, mgctx->ss_mask) {
2320 } while_each_subsys_mask();
2324 goto out_release_tset;
2327 if (tset->nr_tasks) {
2328 do_each_subsys_mask(ss, ssid, mgctx->ss_mask) {
2329 if (ssid == failed_ssid)
2331 if (ss->cancel_attach) {
2333 ss->cancel_attach(tset);
2335 } while_each_subsys_mask();
2338 spin_lock_irq(&css_set_lock);
2339 list_splice_init(&tset->dst_csets, &tset->src_csets);
2340 list_for_each_entry_safe(cset, tmp_cset, &tset->src_csets, mg_node) {
2341 list_splice_tail_init(&cset->mg_tasks, &cset->tasks);
2342 list_del_init(&cset->mg_node);
2344 spin_unlock_irq(&css_set_lock);
2347 * Re-initialize the cgroup_taskset structure in case it is reused
2348 * again in another cgroup_migrate_add_task()/cgroup_migrate_execute()
2352 tset->csets = &tset->src_csets;
2357 * cgroup_migrate_vet_dst - verify whether a cgroup can be migration destination
2358 * @dst_cgrp: destination cgroup to test
2360 * On the default hierarchy, except for the mixable, (possible) thread root
2361 * and threaded cgroups, subtree_control must be zero for migration
2362 * destination cgroups with tasks so that child cgroups don't compete
2365 int cgroup_migrate_vet_dst(struct cgroup *dst_cgrp)
2367 /* v1 doesn't have any restriction */
2368 if (!cgroup_on_dfl(dst_cgrp))
2371 /* verify @dst_cgrp can host resources */
2372 if (!cgroup_is_valid_domain(dst_cgrp->dom_cgrp))
2375 /* mixables don't care */
2376 if (cgroup_is_mixable(dst_cgrp))
2380 * If @dst_cgrp is already or can become a thread root or is
2381 * threaded, it doesn't matter.
2383 if (cgroup_can_be_thread_root(dst_cgrp) || cgroup_is_threaded(dst_cgrp))
2386 /* apply no-internal-process constraint */
2387 if (dst_cgrp->subtree_control)
2394 * cgroup_migrate_finish - cleanup after attach
2395 * @mgctx: migration context
2397 * Undo cgroup_migrate_add_src() and cgroup_migrate_prepare_dst(). See
2398 * those functions for details.
2400 void cgroup_migrate_finish(struct cgroup_mgctx *mgctx)
2402 LIST_HEAD(preloaded);
2403 struct css_set *cset, *tmp_cset;
2405 lockdep_assert_held(&cgroup_mutex);
2407 spin_lock_irq(&css_set_lock);
2409 list_splice_tail_init(&mgctx->preloaded_src_csets, &preloaded);
2410 list_splice_tail_init(&mgctx->preloaded_dst_csets, &preloaded);
2412 list_for_each_entry_safe(cset, tmp_cset, &preloaded, mg_preload_node) {
2413 cset->mg_src_cgrp = NULL;
2414 cset->mg_dst_cgrp = NULL;
2415 cset->mg_dst_cset = NULL;
2416 list_del_init(&cset->mg_preload_node);
2417 put_css_set_locked(cset);
2420 spin_unlock_irq(&css_set_lock);
2424 * cgroup_migrate_add_src - add a migration source css_set
2425 * @src_cset: the source css_set to add
2426 * @dst_cgrp: the destination cgroup
2427 * @mgctx: migration context
2429 * Tasks belonging to @src_cset are about to be migrated to @dst_cgrp. Pin
2430 * @src_cset and add it to @mgctx->src_csets, which should later be cleaned
2431 * up by cgroup_migrate_finish().
2433 * This function may be called without holding cgroup_threadgroup_rwsem
2434 * even if the target is a process. Threads may be created and destroyed
2435 * but as long as cgroup_mutex is not dropped, no new css_set can be put
2436 * into play and the preloaded css_sets are guaranteed to cover all
2439 void cgroup_migrate_add_src(struct css_set *src_cset,
2440 struct cgroup *dst_cgrp,
2441 struct cgroup_mgctx *mgctx)
2443 struct cgroup *src_cgrp;
2445 lockdep_assert_held(&cgroup_mutex);
2446 lockdep_assert_held(&css_set_lock);
2449 * If ->dead, @src_set is associated with one or more dead cgroups
2450 * and doesn't contain any migratable tasks. Ignore it early so
2451 * that the rest of migration path doesn't get confused by it.
2456 src_cgrp = cset_cgroup_from_root(src_cset, dst_cgrp->root);
2458 if (!list_empty(&src_cset->mg_preload_node))
2461 WARN_ON(src_cset->mg_src_cgrp);
2462 WARN_ON(src_cset->mg_dst_cgrp);
2463 WARN_ON(!list_empty(&src_cset->mg_tasks));
2464 WARN_ON(!list_empty(&src_cset->mg_node));
2466 src_cset->mg_src_cgrp = src_cgrp;
2467 src_cset->mg_dst_cgrp = dst_cgrp;
2468 get_css_set(src_cset);
2469 list_add_tail(&src_cset->mg_preload_node, &mgctx->preloaded_src_csets);
2473 * cgroup_migrate_prepare_dst - prepare destination css_sets for migration
2474 * @mgctx: migration context
2476 * Tasks are about to be moved and all the source css_sets have been
2477 * preloaded to @mgctx->preloaded_src_csets. This function looks up and
2478 * pins all destination css_sets, links each to its source, and append them
2479 * to @mgctx->preloaded_dst_csets.
2481 * This function must be called after cgroup_migrate_add_src() has been
2482 * called on each migration source css_set. After migration is performed
2483 * using cgroup_migrate(), cgroup_migrate_finish() must be called on
2486 int cgroup_migrate_prepare_dst(struct cgroup_mgctx *mgctx)
2488 struct css_set *src_cset, *tmp_cset;
2490 lockdep_assert_held(&cgroup_mutex);
2492 /* look up the dst cset for each src cset and link it to src */
2493 list_for_each_entry_safe(src_cset, tmp_cset, &mgctx->preloaded_src_csets,
2495 struct css_set *dst_cset;
2496 struct cgroup_subsys *ss;
2499 dst_cset = find_css_set(src_cset, src_cset->mg_dst_cgrp);
2503 WARN_ON_ONCE(src_cset->mg_dst_cset || dst_cset->mg_dst_cset);
2506 * If src cset equals dst, it's noop. Drop the src.
2507 * cgroup_migrate() will skip the cset too. Note that we
2508 * can't handle src == dst as some nodes are used by both.
2510 if (src_cset == dst_cset) {
2511 src_cset->mg_src_cgrp = NULL;
2512 src_cset->mg_dst_cgrp = NULL;
2513 list_del_init(&src_cset->mg_preload_node);
2514 put_css_set(src_cset);
2515 put_css_set(dst_cset);
2519 src_cset->mg_dst_cset = dst_cset;
2521 if (list_empty(&dst_cset->mg_preload_node))
2522 list_add_tail(&dst_cset->mg_preload_node,
2523 &mgctx->preloaded_dst_csets);
2525 put_css_set(dst_cset);
2527 for_each_subsys(ss, ssid)
2528 if (src_cset->subsys[ssid] != dst_cset->subsys[ssid])
2529 mgctx->ss_mask |= 1 << ssid;
2534 cgroup_migrate_finish(mgctx);
2539 * cgroup_migrate - migrate a process or task to a cgroup
2540 * @leader: the leader of the process or the task to migrate
2541 * @threadgroup: whether @leader points to the whole process or a single task
2542 * @mgctx: migration context
2544 * Migrate a process or task denoted by @leader. If migrating a process,
2545 * the caller must be holding cgroup_threadgroup_rwsem. The caller is also
2546 * responsible for invoking cgroup_migrate_add_src() and
2547 * cgroup_migrate_prepare_dst() on the targets before invoking this
2548 * function and following up with cgroup_migrate_finish().
2550 * As long as a controller's ->can_attach() doesn't fail, this function is
2551 * guaranteed to succeed. This means that, excluding ->can_attach()
2552 * failure, when migrating multiple targets, the success or failure can be
2553 * decided for all targets by invoking group_migrate_prepare_dst() before
2554 * actually starting migrating.
2556 int cgroup_migrate(struct task_struct *leader, bool threadgroup,
2557 struct cgroup_mgctx *mgctx)
2559 struct task_struct *task;
2562 * Prevent freeing of tasks while we take a snapshot. Tasks that are
2563 * already PF_EXITING could be freed from underneath us unless we
2564 * take an rcu_read_lock.
2566 spin_lock_irq(&css_set_lock);
2570 cgroup_migrate_add_task(task, mgctx);
2573 } while_each_thread(leader, task);
2575 spin_unlock_irq(&css_set_lock);
2577 return cgroup_migrate_execute(mgctx);
2581 * cgroup_attach_task - attach a task or a whole threadgroup to a cgroup
2582 * @dst_cgrp: the cgroup to attach to
2583 * @leader: the task or the leader of the threadgroup to be attached
2584 * @threadgroup: attach the whole threadgroup?
2586 * Call holding cgroup_mutex and cgroup_threadgroup_rwsem.
2588 int cgroup_attach_task(struct cgroup *dst_cgrp, struct task_struct *leader,
2591 DEFINE_CGROUP_MGCTX(mgctx);
2592 struct task_struct *task;
2595 ret = cgroup_migrate_vet_dst(dst_cgrp);
2599 /* look up all src csets */
2600 spin_lock_irq(&css_set_lock);
2604 cgroup_migrate_add_src(task_css_set(task), dst_cgrp, &mgctx);
2607 } while_each_thread(leader, task);
2609 spin_unlock_irq(&css_set_lock);
2611 /* prepare dst csets and commit */
2612 ret = cgroup_migrate_prepare_dst(&mgctx);
2614 ret = cgroup_migrate(leader, threadgroup, &mgctx);
2616 cgroup_migrate_finish(&mgctx);
2619 trace_cgroup_attach_task(dst_cgrp, leader, threadgroup);
2624 struct task_struct *cgroup_procs_write_start(char *buf, bool threadgroup)
2625 __acquires(&cgroup_threadgroup_rwsem)
2627 struct task_struct *tsk;
2630 if (kstrtoint(strstrip(buf), 0, &pid) || pid < 0)
2631 return ERR_PTR(-EINVAL);
2633 percpu_down_write(&cgroup_threadgroup_rwsem);
2637 tsk = find_task_by_vpid(pid);
2639 tsk = ERR_PTR(-ESRCH);
2640 goto out_unlock_threadgroup;
2647 tsk = tsk->group_leader;
2650 * kthreads may acquire PF_NO_SETAFFINITY during initialization.
2651 * If userland migrates such a kthread to a non-root cgroup, it can
2652 * become trapped in a cpuset, or RT kthread may be born in a
2653 * cgroup with no rt_runtime allocated. Just say no.
2655 if (tsk->no_cgroup_migration || (tsk->flags & PF_NO_SETAFFINITY)) {
2656 tsk = ERR_PTR(-EINVAL);
2657 goto out_unlock_threadgroup;
2660 get_task_struct(tsk);
2661 goto out_unlock_rcu;
2663 out_unlock_threadgroup:
2664 percpu_up_write(&cgroup_threadgroup_rwsem);
2670 void cgroup_procs_write_finish(struct task_struct *task)
2671 __releases(&cgroup_threadgroup_rwsem)
2673 struct cgroup_subsys *ss;
2676 /* release reference from cgroup_procs_write_start() */
2677 put_task_struct(task);
2679 percpu_up_write(&cgroup_threadgroup_rwsem);
2680 for_each_subsys(ss, ssid)
2681 if (ss->post_attach)
2685 static void cgroup_print_ss_mask(struct seq_file *seq, u16 ss_mask)
2687 struct cgroup_subsys *ss;
2688 bool printed = false;
2691 do_each_subsys_mask(ss, ssid, ss_mask) {
2694 seq_printf(seq, "%s", ss->name);
2696 } while_each_subsys_mask();
2698 seq_putc(seq, '\n');
2701 /* show controllers which are enabled from the parent */
2702 static int cgroup_controllers_show(struct seq_file *seq, void *v)
2704 struct cgroup *cgrp = seq_css(seq)->cgroup;
2706 cgroup_print_ss_mask(seq, cgroup_control(cgrp));
2710 /* show controllers which are enabled for a given cgroup's children */
2711 static int cgroup_subtree_control_show(struct seq_file *seq, void *v)
2713 struct cgroup *cgrp = seq_css(seq)->cgroup;
2715 cgroup_print_ss_mask(seq, cgrp->subtree_control);
2720 * cgroup_update_dfl_csses - update css assoc of a subtree in default hierarchy
2721 * @cgrp: root of the subtree to update csses for
2723 * @cgrp's control masks have changed and its subtree's css associations
2724 * need to be updated accordingly. This function looks up all css_sets
2725 * which are attached to the subtree, creates the matching updated css_sets
2726 * and migrates the tasks to the new ones.
2728 static int cgroup_update_dfl_csses(struct cgroup *cgrp)
2730 DEFINE_CGROUP_MGCTX(mgctx);
2731 struct cgroup_subsys_state *d_css;
2732 struct cgroup *dsct;
2733 struct css_set *src_cset;
2736 lockdep_assert_held(&cgroup_mutex);
2738 percpu_down_write(&cgroup_threadgroup_rwsem);
2740 /* look up all csses currently attached to @cgrp's subtree */
2741 spin_lock_irq(&css_set_lock);
2742 cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) {
2743 struct cgrp_cset_link *link;
2745 list_for_each_entry(link, &dsct->cset_links, cset_link)
2746 cgroup_migrate_add_src(link->cset, dsct, &mgctx);
2748 spin_unlock_irq(&css_set_lock);
2750 /* NULL dst indicates self on default hierarchy */
2751 ret = cgroup_migrate_prepare_dst(&mgctx);
2755 spin_lock_irq(&css_set_lock);
2756 list_for_each_entry(src_cset, &mgctx.preloaded_src_csets, mg_preload_node) {
2757 struct task_struct *task, *ntask;
2759 /* all tasks in src_csets need to be migrated */
2760 list_for_each_entry_safe(task, ntask, &src_cset->tasks, cg_list)
2761 cgroup_migrate_add_task(task, &mgctx);
2763 spin_unlock_irq(&css_set_lock);
2765 ret = cgroup_migrate_execute(&mgctx);
2767 cgroup_migrate_finish(&mgctx);
2768 percpu_up_write(&cgroup_threadgroup_rwsem);
2773 * cgroup_lock_and_drain_offline - lock cgroup_mutex and drain offlined csses
2774 * @cgrp: root of the target subtree
2776 * Because css offlining is asynchronous, userland may try to re-enable a
2777 * controller while the previous css is still around. This function grabs
2778 * cgroup_mutex and drains the previous css instances of @cgrp's subtree.
2780 void cgroup_lock_and_drain_offline(struct cgroup *cgrp)
2781 __acquires(&cgroup_mutex)
2783 struct cgroup *dsct;
2784 struct cgroup_subsys_state *d_css;
2785 struct cgroup_subsys *ss;
2789 mutex_lock(&cgroup_mutex);
2791 cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) {
2792 for_each_subsys(ss, ssid) {
2793 struct cgroup_subsys_state *css = cgroup_css(dsct, ss);
2796 if (!css || !percpu_ref_is_dying(&css->refcnt))
2799 cgroup_get_live(dsct);
2800 prepare_to_wait(&dsct->offline_waitq, &wait,
2801 TASK_UNINTERRUPTIBLE);
2803 mutex_unlock(&cgroup_mutex);
2805 finish_wait(&dsct->offline_waitq, &wait);
2814 * cgroup_save_control - save control masks and dom_cgrp of a subtree
2815 * @cgrp: root of the target subtree
2817 * Save ->subtree_control, ->subtree_ss_mask and ->dom_cgrp to the
2818 * respective old_ prefixed fields for @cgrp's subtree including @cgrp
2821 static void cgroup_save_control(struct cgroup *cgrp)
2823 struct cgroup *dsct;
2824 struct cgroup_subsys_state *d_css;
2826 cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) {
2827 dsct->old_subtree_control = dsct->subtree_control;
2828 dsct->old_subtree_ss_mask = dsct->subtree_ss_mask;
2829 dsct->old_dom_cgrp = dsct->dom_cgrp;
2834 * cgroup_propagate_control - refresh control masks of a subtree
2835 * @cgrp: root of the target subtree
2837 * For @cgrp and its subtree, ensure ->subtree_ss_mask matches
2838 * ->subtree_control and propagate controller availability through the
2839 * subtree so that descendants don't have unavailable controllers enabled.
2841 static void cgroup_propagate_control(struct cgroup *cgrp)
2843 struct cgroup *dsct;
2844 struct cgroup_subsys_state *d_css;
2846 cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) {
2847 dsct->subtree_control &= cgroup_control(dsct);
2848 dsct->subtree_ss_mask =
2849 cgroup_calc_subtree_ss_mask(dsct->subtree_control,
2850 cgroup_ss_mask(dsct));
2855 * cgroup_restore_control - restore control masks and dom_cgrp of a subtree
2856 * @cgrp: root of the target subtree
2858 * Restore ->subtree_control, ->subtree_ss_mask and ->dom_cgrp from the
2859 * respective old_ prefixed fields for @cgrp's subtree including @cgrp
2862 static void cgroup_restore_control(struct cgroup *cgrp)
2864 struct cgroup *dsct;
2865 struct cgroup_subsys_state *d_css;
2867 cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) {
2868 dsct->subtree_control = dsct->old_subtree_control;
2869 dsct->subtree_ss_mask = dsct->old_subtree_ss_mask;
2870 dsct->dom_cgrp = dsct->old_dom_cgrp;
2874 static bool css_visible(struct cgroup_subsys_state *css)
2876 struct cgroup_subsys *ss = css->ss;
2877 struct cgroup *cgrp = css->cgroup;
2879 if (cgroup_control(cgrp) & (1 << ss->id))
2881 if (!(cgroup_ss_mask(cgrp) & (1 << ss->id)))
2883 return cgroup_on_dfl(cgrp) && ss->implicit_on_dfl;
2887 * cgroup_apply_control_enable - enable or show csses according to control
2888 * @cgrp: root of the target subtree
2890 * Walk @cgrp's subtree and create new csses or make the existing ones
2891 * visible. A css is created invisible if it's being implicitly enabled
2892 * through dependency. An invisible css is made visible when the userland
2893 * explicitly enables it.
2895 * Returns 0 on success, -errno on failure. On failure, csses which have
2896 * been processed already aren't cleaned up. The caller is responsible for
2897 * cleaning up with cgroup_apply_control_disable().
2899 static int cgroup_apply_control_enable(struct cgroup *cgrp)
2901 struct cgroup *dsct;
2902 struct cgroup_subsys_state *d_css;
2903 struct cgroup_subsys *ss;
2906 cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) {
2907 for_each_subsys(ss, ssid) {
2908 struct cgroup_subsys_state *css = cgroup_css(dsct, ss);
2910 if (!(cgroup_ss_mask(dsct) & (1 << ss->id)))
2914 css = css_create(dsct, ss);
2916 return PTR_ERR(css);
2919 WARN_ON_ONCE(percpu_ref_is_dying(&css->refcnt));
2921 if (css_visible(css)) {
2922 ret = css_populate_dir(css);
2933 * cgroup_apply_control_disable - kill or hide csses according to control
2934 * @cgrp: root of the target subtree
2936 * Walk @cgrp's subtree and kill and hide csses so that they match
2937 * cgroup_ss_mask() and cgroup_visible_mask().
2939 * A css is hidden when the userland requests it to be disabled while other
2940 * subsystems are still depending on it. The css must not actively control
2941 * resources and be in the vanilla state if it's made visible again later.
2942 * Controllers which may be depended upon should provide ->css_reset() for
2945 static void cgroup_apply_control_disable(struct cgroup *cgrp)
2947 struct cgroup *dsct;
2948 struct cgroup_subsys_state *d_css;
2949 struct cgroup_subsys *ss;
2952 cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) {
2953 for_each_subsys(ss, ssid) {
2954 struct cgroup_subsys_state *css = cgroup_css(dsct, ss);
2959 WARN_ON_ONCE(percpu_ref_is_dying(&css->refcnt));
2962 !(cgroup_ss_mask(dsct) & (1 << ss->id))) {
2964 } else if (!css_visible(css)) {
2974 * cgroup_apply_control - apply control mask updates to the subtree
2975 * @cgrp: root of the target subtree
2977 * subsystems can be enabled and disabled in a subtree using the following
2980 * 1. Call cgroup_save_control() to stash the current state.
2981 * 2. Update ->subtree_control masks in the subtree as desired.
2982 * 3. Call cgroup_apply_control() to apply the changes.
2983 * 4. Optionally perform other related operations.
2984 * 5. Call cgroup_finalize_control() to finish up.
2986 * This function implements step 3 and propagates the mask changes
2987 * throughout @cgrp's subtree, updates csses accordingly and perform
2988 * process migrations.
2990 static int cgroup_apply_control(struct cgroup *cgrp)
2994 cgroup_propagate_control(cgrp);
2996 ret = cgroup_apply_control_enable(cgrp);
3001 * At this point, cgroup_e_css() results reflect the new csses
3002 * making the following cgroup_update_dfl_csses() properly update
3003 * css associations of all tasks in the subtree.
3005 ret = cgroup_update_dfl_csses(cgrp);
3013 * cgroup_finalize_control - finalize control mask update
3014 * @cgrp: root of the target subtree
3015 * @ret: the result of the update
3017 * Finalize control mask update. See cgroup_apply_control() for more info.
3019 static void cgroup_finalize_control(struct cgroup *cgrp, int ret)
3022 cgroup_restore_control(cgrp);
3023 cgroup_propagate_control(cgrp);
3026 cgroup_apply_control_disable(cgrp);
3029 static int cgroup_vet_subtree_control_enable(struct cgroup *cgrp, u16 enable)
3031 u16 domain_enable = enable & ~cgrp_dfl_threaded_ss_mask;
3033 /* if nothing is getting enabled, nothing to worry about */
3037 /* can @cgrp host any resources? */
3038 if (!cgroup_is_valid_domain(cgrp->dom_cgrp))
3041 /* mixables don't care */
3042 if (cgroup_is_mixable(cgrp))
3045 if (domain_enable) {
3046 /* can't enable domain controllers inside a thread subtree */
3047 if (cgroup_is_thread_root(cgrp) || cgroup_is_threaded(cgrp))
3051 * Threaded controllers can handle internal competitions
3052 * and are always allowed inside a (prospective) thread
3055 if (cgroup_can_be_thread_root(cgrp) || cgroup_is_threaded(cgrp))
3060 * Controllers can't be enabled for a cgroup with tasks to avoid
3061 * child cgroups competing against tasks.
3063 if (cgroup_has_tasks(cgrp))
3069 /* change the enabled child controllers for a cgroup in the default hierarchy */
3070 static ssize_t cgroup_subtree_control_write(struct kernfs_open_file *of,
3071 char *buf, size_t nbytes,
3074 u16 enable = 0, disable = 0;
3075 struct cgroup *cgrp, *child;
3076 struct cgroup_subsys *ss;
3081 * Parse input - space separated list of subsystem names prefixed
3082 * with either + or -.
3084 buf = strstrip(buf);
3085 while ((tok = strsep(&buf, " "))) {
3088 do_each_subsys_mask(ss, ssid, ~cgrp_dfl_inhibit_ss_mask) {
3089 if (!cgroup_ssid_enabled(ssid) ||
3090 strcmp(tok + 1, ss->name))
3094 enable |= 1 << ssid;
3095 disable &= ~(1 << ssid);
3096 } else if (*tok == '-') {
3097 disable |= 1 << ssid;
3098 enable &= ~(1 << ssid);
3103 } while_each_subsys_mask();
3104 if (ssid == CGROUP_SUBSYS_COUNT)
3108 cgrp = cgroup_kn_lock_live(of->kn, true);
3112 for_each_subsys(ss, ssid) {
3113 if (enable & (1 << ssid)) {
3114 if (cgrp->subtree_control & (1 << ssid)) {
3115 enable &= ~(1 << ssid);
3119 if (!(cgroup_control(cgrp) & (1 << ssid))) {
3123 } else if (disable & (1 << ssid)) {
3124 if (!(cgrp->subtree_control & (1 << ssid))) {
3125 disable &= ~(1 << ssid);
3129 /* a child has it enabled? */
3130 cgroup_for_each_live_child(child, cgrp) {
3131 if (child->subtree_control & (1 << ssid)) {
3139 if (!enable && !disable) {
3144 ret = cgroup_vet_subtree_control_enable(cgrp, enable);
3148 /* save and update control masks and prepare csses */
3149 cgroup_save_control(cgrp);
3151 cgrp->subtree_control |= enable;
3152 cgrp->subtree_control &= ~disable;
3154 ret = cgroup_apply_control(cgrp);
3155 cgroup_finalize_control(cgrp, ret);
3159 kernfs_activate(cgrp->kn);
3161 cgroup_kn_unlock(of->kn);
3162 return ret ?: nbytes;
3166 * cgroup_enable_threaded - make @cgrp threaded
3167 * @cgrp: the target cgroup
3169 * Called when "threaded" is written to the cgroup.type interface file and
3170 * tries to make @cgrp threaded and join the parent's resource domain.
3171 * This function is never called on the root cgroup as cgroup.type doesn't
3174 static int cgroup_enable_threaded(struct cgroup *cgrp)
3176 struct cgroup *parent = cgroup_parent(cgrp);
3177 struct cgroup *dom_cgrp = parent->dom_cgrp;
3178 struct cgroup *dsct;
3179 struct cgroup_subsys_state *d_css;
3182 lockdep_assert_held(&cgroup_mutex);
3184 /* noop if already threaded */
3185 if (cgroup_is_threaded(cgrp))
3189 * If @cgroup is populated or has domain controllers enabled, it
3190 * can't be switched. While the below cgroup_can_be_thread_root()
3191 * test can catch the same conditions, that's only when @parent is
3192 * not mixable, so let's check it explicitly.
3194 if (cgroup_is_populated(cgrp) ||
3195 cgrp->subtree_control & ~cgrp_dfl_threaded_ss_mask)
3198 /* we're joining the parent's domain, ensure its validity */
3199 if (!cgroup_is_valid_domain(dom_cgrp) ||
3200 !cgroup_can_be_thread_root(dom_cgrp))
3204 * The following shouldn't cause actual migrations and should
3207 cgroup_save_control(cgrp);
3209 cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp)
3210 if (dsct == cgrp || cgroup_is_threaded(dsct))
3211 dsct->dom_cgrp = dom_cgrp;
3213 ret = cgroup_apply_control(cgrp);
3215 parent->nr_threaded_children++;
3217 cgroup_finalize_control(cgrp, ret);
3221 static int cgroup_type_show(struct seq_file *seq, void *v)
3223 struct cgroup *cgrp = seq_css(seq)->cgroup;
3225 if (cgroup_is_threaded(cgrp))
3226 seq_puts(seq, "threaded\n");
3227 else if (!cgroup_is_valid_domain(cgrp))
3228 seq_puts(seq, "domain invalid\n");
3229 else if (cgroup_is_thread_root(cgrp))
3230 seq_puts(seq, "domain threaded\n");
3232 seq_puts(seq, "domain\n");
3237 static ssize_t cgroup_type_write(struct kernfs_open_file *of, char *buf,
3238 size_t nbytes, loff_t off)
3240 struct cgroup *cgrp;
3243 /* only switching to threaded mode is supported */
3244 if (strcmp(strstrip(buf), "threaded"))
3247 /* drain dying csses before we re-apply (threaded) subtree control */
3248 cgrp = cgroup_kn_lock_live(of->kn, true);
3252 /* threaded can only be enabled */
3253 ret = cgroup_enable_threaded(cgrp);
3255 cgroup_kn_unlock(of->kn);
3256 return ret ?: nbytes;
3259 static int cgroup_max_descendants_show(struct seq_file *seq, void *v)
3261 struct cgroup *cgrp = seq_css(seq)->cgroup;
3262 int descendants = READ_ONCE(cgrp->max_descendants);
3264 if (descendants == INT_MAX)
3265 seq_puts(seq, "max\n");
3267 seq_printf(seq, "%d\n", descendants);
3272 static ssize_t cgroup_max_descendants_write(struct kernfs_open_file *of,
3273 char *buf, size_t nbytes, loff_t off)
3275 struct cgroup *cgrp;
3279 buf = strstrip(buf);
3280 if (!strcmp(buf, "max")) {
3281 descendants = INT_MAX;
3283 ret = kstrtoint(buf, 0, &descendants);
3288 if (descendants < 0)
3291 cgrp = cgroup_kn_lock_live(of->kn, false);
3295 cgrp->max_descendants = descendants;
3297 cgroup_kn_unlock(of->kn);
3302 static int cgroup_max_depth_show(struct seq_file *seq, void *v)
3304 struct cgroup *cgrp = seq_css(seq)->cgroup;
3305 int depth = READ_ONCE(cgrp->max_depth);
3307 if (depth == INT_MAX)
3308 seq_puts(seq, "max\n");
3310 seq_printf(seq, "%d\n", depth);
3315 static ssize_t cgroup_max_depth_write(struct kernfs_open_file *of,
3316 char *buf, size_t nbytes, loff_t off)
3318 struct cgroup *cgrp;
3322 buf = strstrip(buf);
3323 if (!strcmp(buf, "max")) {
3326 ret = kstrtoint(buf, 0, &depth);
3334 cgrp = cgroup_kn_lock_live(of->kn, false);
3338 cgrp->max_depth = depth;
3340 cgroup_kn_unlock(of->kn);
3345 static int cgroup_events_show(struct seq_file *seq, void *v)
3347 seq_printf(seq, "populated %d\n",
3348 cgroup_is_populated(seq_css(seq)->cgroup));
3352 static int cgroup_stat_show(struct seq_file *seq, void *v)
3354 struct cgroup *cgroup = seq_css(seq)->cgroup;
3356 seq_printf(seq, "nr_descendants %d\n",
3357 cgroup->nr_descendants);
3358 seq_printf(seq, "nr_dying_descendants %d\n",
3359 cgroup->nr_dying_descendants);
3364 static int cgroup_file_open(struct kernfs_open_file *of)
3366 struct cftype *cft = of->kn->priv;
3369 return cft->open(of);
3373 static void cgroup_file_release(struct kernfs_open_file *of)
3375 struct cftype *cft = of->kn->priv;
3381 static ssize_t cgroup_file_write(struct kernfs_open_file *of, char *buf,
3382 size_t nbytes, loff_t off)
3384 struct cgroup_namespace *ns = current->nsproxy->cgroup_ns;
3385 struct cgroup *cgrp = of->kn->parent->priv;
3386 struct cftype *cft = of->kn->priv;
3387 struct cgroup_subsys_state *css;
3391 * If namespaces are delegation boundaries, disallow writes to
3392 * files in an non-init namespace root from inside the namespace
3393 * except for the files explicitly marked delegatable -
3394 * cgroup.procs and cgroup.subtree_control.
3396 if ((cgrp->root->flags & CGRP_ROOT_NS_DELEGATE) &&
3397 !(cft->flags & CFTYPE_NS_DELEGATABLE) &&
3398 ns != &init_cgroup_ns && ns->root_cset->dfl_cgrp == cgrp)
3402 return cft->write(of, buf, nbytes, off);
3405 * kernfs guarantees that a file isn't deleted with operations in
3406 * flight, which means that the matching css is and stays alive and
3407 * doesn't need to be pinned. The RCU locking is not necessary
3408 * either. It's just for the convenience of using cgroup_css().
3411 css = cgroup_css(cgrp, cft->ss);
3414 if (cft->write_u64) {
3415 unsigned long long v;
3416 ret = kstrtoull(buf, 0, &v);
3418 ret = cft->write_u64(css, cft, v);
3419 } else if (cft->write_s64) {
3421 ret = kstrtoll(buf, 0, &v);
3423 ret = cft->write_s64(css, cft, v);
3428 return ret ?: nbytes;
3431 static void *cgroup_seqfile_start(struct seq_file *seq, loff_t *ppos)
3433 return seq_cft(seq)->seq_start(seq, ppos);
3436 static void *cgroup_seqfile_next(struct seq_file *seq, void *v, loff_t *ppos)
3438 return seq_cft(seq)->seq_next(seq, v, ppos);
3441 static void cgroup_seqfile_stop(struct seq_file *seq, void *v)
3443 if (seq_cft(seq)->seq_stop)
3444 seq_cft(seq)->seq_stop(seq, v);
3447 static int cgroup_seqfile_show(struct seq_file *m, void *arg)
3449 struct cftype *cft = seq_cft(m);
3450 struct cgroup_subsys_state *css = seq_css(m);
3453 return cft->seq_show(m, arg);
3456 seq_printf(m, "%llu\n", cft->read_u64(css, cft));
3457 else if (cft->read_s64)
3458 seq_printf(m, "%lld\n", cft->read_s64(css, cft));
3464 static struct kernfs_ops cgroup_kf_single_ops = {
3465 .atomic_write_len = PAGE_SIZE,
3466 .open = cgroup_file_open,
3467 .release = cgroup_file_release,
3468 .write = cgroup_file_write,
3469 .seq_show = cgroup_seqfile_show,
3472 static struct kernfs_ops cgroup_kf_ops = {
3473 .atomic_write_len = PAGE_SIZE,
3474 .open = cgroup_file_open,
3475 .release = cgroup_file_release,
3476 .write = cgroup_file_write,
3477 .seq_start = cgroup_seqfile_start,
3478 .seq_next = cgroup_seqfile_next,
3479 .seq_stop = cgroup_seqfile_stop,
3480 .seq_show = cgroup_seqfile_show,
3483 /* set uid and gid of cgroup dirs and files to that of the creator */
3484 static int cgroup_kn_set_ugid(struct kernfs_node *kn)
3486 struct iattr iattr = { .ia_valid = ATTR_UID | ATTR_GID,
3487 .ia_uid = current_fsuid(),
3488 .ia_gid = current_fsgid(), };
3490 if (uid_eq(iattr.ia_uid, GLOBAL_ROOT_UID) &&
3491 gid_eq(iattr.ia_gid, GLOBAL_ROOT_GID))
3494 return kernfs_setattr(kn, &iattr);
3497 static int cgroup_add_file(struct cgroup_subsys_state *css, struct cgroup *cgrp,
3500 char name[CGROUP_FILE_NAME_MAX];
3501 struct kernfs_node *kn;
3502 struct lock_class_key *key = NULL;
3505 #ifdef CONFIG_DEBUG_LOCK_ALLOC
3506 key = &cft->lockdep_key;
3508 kn = __kernfs_create_file(cgrp->kn, cgroup_file_name(cgrp, cft, name),
3509 cgroup_file_mode(cft), 0, cft->kf_ops, cft,
3514 ret = cgroup_kn_set_ugid(kn);
3520 if (cft->file_offset) {
3521 struct cgroup_file *cfile = (void *)css + cft->file_offset;
3523 spin_lock_irq(&cgroup_file_kn_lock);
3525 spin_unlock_irq(&cgroup_file_kn_lock);
3532 * cgroup_addrm_files - add or remove files to a cgroup directory
3533 * @css: the target css
3534 * @cgrp: the target cgroup (usually css->cgroup)
3535 * @cfts: array of cftypes to be added
3536 * @is_add: whether to add or remove
3538 * Depending on @is_add, add or remove files defined by @cfts on @cgrp.
3539 * For removals, this function never fails.
3541 static int cgroup_addrm_files(struct cgroup_subsys_state *css,
3542 struct cgroup *cgrp, struct cftype cfts[],
3545 struct cftype *cft, *cft_end = NULL;
3548 lockdep_assert_held(&cgroup_mutex);
3551 for (cft = cfts; cft != cft_end && cft->name[0] != '\0'; cft++) {
3552 /* does cft->flags tell us to skip this file on @cgrp? */
3553 if ((cft->flags & __CFTYPE_ONLY_ON_DFL) && !cgroup_on_dfl(cgrp))
3555 if ((cft->flags & __CFTYPE_NOT_ON_DFL) && cgroup_on_dfl(cgrp))
3557 if ((cft->flags & CFTYPE_NOT_ON_ROOT) && !cgroup_parent(cgrp))
3559 if ((cft->flags & CFTYPE_ONLY_ON_ROOT) && cgroup_parent(cgrp))
3563 ret = cgroup_add_file(css, cgrp, cft);
3565 pr_warn("%s: failed to add %s, err=%d\n",
3566 __func__, cft->name, ret);
3572 cgroup_rm_file(cgrp, cft);
3578 static int cgroup_apply_cftypes(struct cftype *cfts, bool is_add)
3580 struct cgroup_subsys *ss = cfts[0].ss;
3581 struct cgroup *root = &ss->root->cgrp;
3582 struct cgroup_subsys_state *css;
3585 lockdep_assert_held(&cgroup_mutex);
3587 /* add/rm files for all cgroups created before */
3588 css_for_each_descendant_pre(css, cgroup_css(root, ss)) {
3589 struct cgroup *cgrp = css->cgroup;
3591 if (!(css->flags & CSS_VISIBLE))
3594 ret = cgroup_addrm_files(css, cgrp, cfts, is_add);
3600 kernfs_activate(root->kn);
3604 static void cgroup_exit_cftypes(struct cftype *cfts)
3608 for (cft = cfts; cft->name[0] != '\0'; cft++) {
3609 /* free copy for custom atomic_write_len, see init_cftypes() */
3610 if (cft->max_write_len && cft->max_write_len != PAGE_SIZE)
3615 /* revert flags set by cgroup core while adding @cfts */
3616 cft->flags &= ~(__CFTYPE_ONLY_ON_DFL | __CFTYPE_NOT_ON_DFL);
3620 static int cgroup_init_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
3624 for (cft = cfts; cft->name[0] != '\0'; cft++) {
3625 struct kernfs_ops *kf_ops;
3627 WARN_ON(cft->ss || cft->kf_ops);
3630 kf_ops = &cgroup_kf_ops;
3632 kf_ops = &cgroup_kf_single_ops;
3635 * Ugh... if @cft wants a custom max_write_len, we need to
3636 * make a copy of kf_ops to set its atomic_write_len.
3638 if (cft->max_write_len && cft->max_write_len != PAGE_SIZE) {
3639 kf_ops = kmemdup(kf_ops, sizeof(*kf_ops), GFP_KERNEL);
3641 cgroup_exit_cftypes(cfts);
3644 kf_ops->atomic_write_len = cft->max_write_len;
3647 cft->kf_ops = kf_ops;
3654 static int cgroup_rm_cftypes_locked(struct cftype *cfts)
3656 lockdep_assert_held(&cgroup_mutex);
3658 if (!cfts || !cfts[0].ss)
3661 list_del(&cfts->node);
3662 cgroup_apply_cftypes(cfts, false);
3663 cgroup_exit_cftypes(cfts);
3668 * cgroup_rm_cftypes - remove an array of cftypes from a subsystem
3669 * @cfts: zero-length name terminated array of cftypes
3671 * Unregister @cfts. Files described by @cfts are removed from all
3672 * existing cgroups and all future cgroups won't have them either. This
3673 * function can be called anytime whether @cfts' subsys is attached or not.
3675 * Returns 0 on successful unregistration, -ENOENT if @cfts is not
3678 int cgroup_rm_cftypes(struct cftype *cfts)
3682 mutex_lock(&cgroup_mutex);
3683 ret = cgroup_rm_cftypes_locked(cfts);
3684 mutex_unlock(&cgroup_mutex);
3689 * cgroup_add_cftypes - add an array of cftypes to a subsystem
3690 * @ss: target cgroup subsystem
3691 * @cfts: zero-length name terminated array of cftypes
3693 * Register @cfts to @ss. Files described by @cfts are created for all
3694 * existing cgroups to which @ss is attached and all future cgroups will
3695 * have them too. This function can be called anytime whether @ss is
3698 * Returns 0 on successful registration, -errno on failure. Note that this
3699 * function currently returns 0 as long as @cfts registration is successful
3700 * even if some file creation attempts on existing cgroups fail.
3702 static int cgroup_add_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
3706 if (!cgroup_ssid_enabled(ss->id))
3709 if (!cfts || cfts[0].name[0] == '\0')
3712 ret = cgroup_init_cftypes(ss, cfts);
3716 mutex_lock(&cgroup_mutex);
3718 list_add_tail(&cfts->node, &ss->cfts);
3719 ret = cgroup_apply_cftypes(cfts, true);
3721 cgroup_rm_cftypes_locked(cfts);
3723 mutex_unlock(&cgroup_mutex);
3728 * cgroup_add_dfl_cftypes - add an array of cftypes for default hierarchy
3729 * @ss: target cgroup subsystem
3730 * @cfts: zero-length name terminated array of cftypes
3732 * Similar to cgroup_add_cftypes() but the added files are only used for
3733 * the default hierarchy.
3735 int cgroup_add_dfl_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
3739 for (cft = cfts; cft && cft->name[0] != '\0'; cft++)
3740 cft->flags |= __CFTYPE_ONLY_ON_DFL;
3741 return cgroup_add_cftypes(ss, cfts);
3745 * cgroup_add_legacy_cftypes - add an array of cftypes for legacy hierarchies
3746 * @ss: target cgroup subsystem
3747 * @cfts: zero-length name terminated array of cftypes
3749 * Similar to cgroup_add_cftypes() but the added files are only used for
3750 * the legacy hierarchies.
3752 int cgroup_add_legacy_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
3756 for (cft = cfts; cft && cft->name[0] != '\0'; cft++)
3757 cft->flags |= __CFTYPE_NOT_ON_DFL;
3758 return cgroup_add_cftypes(ss, cfts);
3762 * cgroup_file_notify - generate a file modified event for a cgroup_file
3763 * @cfile: target cgroup_file
3765 * @cfile must have been obtained by setting cftype->file_offset.
3767 void cgroup_file_notify(struct cgroup_file *cfile)
3769 unsigned long flags;
3771 spin_lock_irqsave(&cgroup_file_kn_lock, flags);
3773 kernfs_notify(cfile->kn);
3774 spin_unlock_irqrestore(&cgroup_file_kn_lock, flags);
3778 * css_next_child - find the next child of a given css
3779 * @pos: the current position (%NULL to initiate traversal)
3780 * @parent: css whose children to walk
3782 * This function returns the next child of @parent and should be called
3783 * under either cgroup_mutex or RCU read lock. The only requirement is
3784 * that @parent and @pos are accessible. The next sibling is guaranteed to
3785 * be returned regardless of their states.
3787 * If a subsystem synchronizes ->css_online() and the start of iteration, a
3788 * css which finished ->css_online() is guaranteed to be visible in the
3789 * future iterations and will stay visible until the last reference is put.
3790 * A css which hasn't finished ->css_online() or already finished
3791 * ->css_offline() may show up during traversal. It's each subsystem's
3792 * responsibility to synchronize against on/offlining.
3794 struct cgroup_subsys_state *css_next_child(struct cgroup_subsys_state *pos,
3795 struct cgroup_subsys_state *parent)
3797 struct cgroup_subsys_state *next;
3799 cgroup_assert_mutex_or_rcu_locked();
3802 * @pos could already have been unlinked from the sibling list.
3803 * Once a cgroup is removed, its ->sibling.next is no longer
3804 * updated when its next sibling changes. CSS_RELEASED is set when
3805 * @pos is taken off list, at which time its next pointer is valid,
3806 * and, as releases are serialized, the one pointed to by the next
3807 * pointer is guaranteed to not have started release yet. This
3808 * implies that if we observe !CSS_RELEASED on @pos in this RCU
3809 * critical section, the one pointed to by its next pointer is
3810 * guaranteed to not have finished its RCU grace period even if we
3811 * have dropped rcu_read_lock() inbetween iterations.
3813 * If @pos has CSS_RELEASED set, its next pointer can't be
3814 * dereferenced; however, as each css is given a monotonically
3815 * increasing unique serial number and always appended to the
3816 * sibling list, the next one can be found by walking the parent's
3817 * children until the first css with higher serial number than
3818 * @pos's. While this path can be slower, it happens iff iteration
3819 * races against release and the race window is very small.
3822 next = list_entry_rcu(parent->children.next, struct cgroup_subsys_state, sibling);
3823 } else if (likely(!(pos->flags & CSS_RELEASED))) {
3824 next = list_entry_rcu(pos->sibling.next, struct cgroup_subsys_state, sibling);
3826 list_for_each_entry_rcu(next, &parent->children, sibling)
3827 if (next->serial_nr > pos->serial_nr)
3832 * @next, if not pointing to the head, can be dereferenced and is
3835 if (&next->sibling != &parent->children)
3841 * css_next_descendant_pre - find the next descendant for pre-order walk
3842 * @pos: the current position (%NULL to initiate traversal)
3843 * @root: css whose descendants to walk
3845 * To be used by css_for_each_descendant_pre(). Find the next descendant
3846 * to visit for pre-order traversal of @root's descendants. @root is
3847 * included in the iteration and the first node to be visited.
3849 * While this function requires cgroup_mutex or RCU read locking, it
3850 * doesn't require the whole traversal to be contained in a single critical
3851 * section. This function will return the correct next descendant as long
3852 * as both @pos and @root are accessible and @pos is a descendant of @root.
3854 * If a subsystem synchronizes ->css_online() and the start of iteration, a
3855 * css which finished ->css_online() is guaranteed to be visible in the
3856 * future iterations and will stay visible until the last reference is put.
3857 * A css which hasn't finished ->css_online() or already finished
3858 * ->css_offline() may show up during traversal. It's each subsystem's
3859 * responsibility to synchronize against on/offlining.
3861 struct cgroup_subsys_state *
3862 css_next_descendant_pre(struct cgroup_subsys_state *pos,
3863 struct cgroup_subsys_state *root)
3865 struct cgroup_subsys_state *next;
3867 cgroup_assert_mutex_or_rcu_locked();
3869 /* if first iteration, visit @root */
3873 /* visit the first child if exists */
3874 next = css_next_child(NULL, pos);
3878 /* no child, visit my or the closest ancestor's next sibling */
3879 while (pos != root) {
3880 next = css_next_child(pos, pos->parent);
3890 * css_rightmost_descendant - return the rightmost descendant of a css
3891 * @pos: css of interest
3893 * Return the rightmost descendant of @pos. If there's no descendant, @pos
3894 * is returned. This can be used during pre-order traversal to skip
3897 * While this function requires cgroup_mutex or RCU read locking, it
3898 * doesn't require the whole traversal to be contained in a single critical
3899 * section. This function will return the correct rightmost descendant as
3900 * long as @pos is accessible.
3902 struct cgroup_subsys_state *
3903 css_rightmost_descendant(struct cgroup_subsys_state *pos)
3905 struct cgroup_subsys_state *last, *tmp;
3907 cgroup_assert_mutex_or_rcu_locked();
3911 /* ->prev isn't RCU safe, walk ->next till the end */
3913 css_for_each_child(tmp, last)
3920 static struct cgroup_subsys_state *
3921 css_leftmost_descendant(struct cgroup_subsys_state *pos)
3923 struct cgroup_subsys_state *last;
3927 pos = css_next_child(NULL, pos);
3934 * css_next_descendant_post - find the next descendant for post-order walk
3935 * @pos: the current position (%NULL to initiate traversal)
3936 * @root: css whose descendants to walk
3938 * To be used by css_for_each_descendant_post(). Find the next descendant
3939 * to visit for post-order traversal of @root's descendants. @root is
3940 * included in the iteration and the last node to be visited.
3942 * While this function requires cgroup_mutex or RCU read locking, it
3943 * doesn't require the whole traversal to be contained in a single critical
3944 * section. This function will return the correct next descendant as long
3945 * as both @pos and @cgroup are accessible and @pos is a descendant of
3948 * If a subsystem synchronizes ->css_online() and the start of iteration, a
3949 * css which finished ->css_online() is guaranteed to be visible in the
3950 * future iterations and will stay visible until the last reference is put.
3951 * A css which hasn't finished ->css_online() or already finished
3952 * ->css_offline() may show up during traversal. It's each subsystem's
3953 * responsibility to synchronize against on/offlining.
3955 struct cgroup_subsys_state *
3956 css_next_descendant_post(struct cgroup_subsys_state *pos,
3957 struct cgroup_subsys_state *root)
3959 struct cgroup_subsys_state *next;
3961 cgroup_assert_mutex_or_rcu_locked();
3963 /* if first iteration, visit leftmost descendant which may be @root */
3965 return css_leftmost_descendant(root);
3967 /* if we visited @root, we're done */
3971 /* if there's an unvisited sibling, visit its leftmost descendant */
3972 next = css_next_child(pos, pos->parent);
3974 return css_leftmost_descendant(next);
3976 /* no sibling left, visit parent */
3981 * css_has_online_children - does a css have online children
3982 * @css: the target css
3984 * Returns %true if @css has any online children; otherwise, %false. This
3985 * function can be called from any context but the caller is responsible
3986 * for synchronizing against on/offlining as necessary.
3988 bool css_has_online_children(struct cgroup_subsys_state *css)
3990 struct cgroup_subsys_state *child;
3994 css_for_each_child(child, css) {
3995 if (child->flags & CSS_ONLINE) {
4004 static struct css_set *css_task_iter_next_css_set(struct css_task_iter *it)
4006 struct list_head *l;
4007 struct cgrp_cset_link *link;
4008 struct css_set *cset;
4010 lockdep_assert_held(&css_set_lock);
4012 /* find the next threaded cset */
4013 if (it->tcset_pos) {
4014 l = it->tcset_pos->next;
4016 if (l != it->tcset_head) {
4018 return container_of(l, struct css_set,
4019 threaded_csets_node);
4022 it->tcset_pos = NULL;
4025 /* find the next cset */
4028 if (l == it->cset_head) {
4029 it->cset_pos = NULL;
4034 cset = container_of(l, struct css_set, e_cset_node[it->ss->id]);
4036 link = list_entry(l, struct cgrp_cset_link, cset_link);
4042 /* initialize threaded css_set walking */
4043 if (it->flags & CSS_TASK_ITER_THREADED) {
4045 put_css_set_locked(it->cur_dcset);
4046 it->cur_dcset = cset;
4049 it->tcset_head = &cset->threaded_csets;
4050 it->tcset_pos = &cset->threaded_csets;
4057 * css_task_iter_advance_css_set - advance a task itererator to the next css_set
4058 * @it: the iterator to advance
4060 * Advance @it to the next css_set to walk.
4062 static void css_task_iter_advance_css_set(struct css_task_iter *it)
4064 struct css_set *cset;
4066 lockdep_assert_held(&css_set_lock);
4068 /* Advance to the next non-empty css_set */
4070 cset = css_task_iter_next_css_set(it);
4072 it->task_pos = NULL;
4075 } while (!css_set_populated(cset) && list_empty(&cset->dying_tasks));
4077 if (!list_empty(&cset->tasks)) {
4078 it->task_pos = cset->tasks.next;
4079 it->cur_tasks_head = &cset->tasks;
4080 } else if (!list_empty(&cset->mg_tasks)) {
4081 it->task_pos = cset->mg_tasks.next;
4082 it->cur_tasks_head = &cset->mg_tasks;
4084 it->task_pos = cset->dying_tasks.next;
4085 it->cur_tasks_head = &cset->dying_tasks;
4088 it->tasks_head = &cset->tasks;
4089 it->mg_tasks_head = &cset->mg_tasks;
4090 it->dying_tasks_head = &cset->dying_tasks;
4093 * We don't keep css_sets locked across iteration steps and thus
4094 * need to take steps to ensure that iteration can be resumed after
4095 * the lock is re-acquired. Iteration is performed at two levels -
4096 * css_sets and tasks in them.
4098 * Once created, a css_set never leaves its cgroup lists, so a
4099 * pinned css_set is guaranteed to stay put and we can resume
4100 * iteration afterwards.
4102 * Tasks may leave @cset across iteration steps. This is resolved
4103 * by registering each iterator with the css_set currently being
4104 * walked and making css_set_move_task() advance iterators whose
4105 * next task is leaving.
4108 list_del(&it->iters_node);
4109 put_css_set_locked(it->cur_cset);
4112 it->cur_cset = cset;
4113 list_add(&it->iters_node, &cset->task_iters);
4116 static void css_task_iter_skip(struct css_task_iter *it,
4117 struct task_struct *task)
4119 lockdep_assert_held(&css_set_lock);
4121 if (it->task_pos == &task->cg_list) {
4122 it->task_pos = it->task_pos->next;
4123 it->flags |= CSS_TASK_ITER_SKIPPED;
4127 static void css_task_iter_advance(struct css_task_iter *it)
4129 struct task_struct *task;
4131 lockdep_assert_held(&css_set_lock);
4135 * Advance iterator to find next entry. cset->tasks is
4136 * consumed first and then ->mg_tasks. After ->mg_tasks,
4137 * we move onto the next cset.
4139 if (it->flags & CSS_TASK_ITER_SKIPPED)
4140 it->flags &= ~CSS_TASK_ITER_SKIPPED;
4142 it->task_pos = it->task_pos->next;
4144 if (it->task_pos == it->tasks_head) {
4145 it->task_pos = it->mg_tasks_head->next;
4146 it->cur_tasks_head = it->mg_tasks_head;
4148 if (it->task_pos == it->mg_tasks_head) {
4149 it->task_pos = it->dying_tasks_head->next;
4150 it->cur_tasks_head = it->dying_tasks_head;
4152 if (it->task_pos == it->dying_tasks_head)
4153 css_task_iter_advance_css_set(it);
4155 /* called from start, proceed to the first cset */
4156 css_task_iter_advance_css_set(it);
4162 task = list_entry(it->task_pos, struct task_struct, cg_list);
4164 if (it->flags & CSS_TASK_ITER_PROCS) {
4165 /* if PROCS, skip over tasks which aren't group leaders */
4166 if (!thread_group_leader(task))
4169 /* and dying leaders w/o live member threads */
4170 if (it->cur_tasks_head == it->dying_tasks_head &&
4171 !atomic_read(&task->signal->live))
4174 /* skip all dying ones */
4175 if (it->cur_tasks_head == it->dying_tasks_head)
4181 * css_task_iter_start - initiate task iteration
4182 * @css: the css to walk tasks of
4183 * @flags: CSS_TASK_ITER_* flags
4184 * @it: the task iterator to use
4186 * Initiate iteration through the tasks of @css. The caller can call
4187 * css_task_iter_next() to walk through the tasks until the function
4188 * returns NULL. On completion of iteration, css_task_iter_end() must be
4191 void css_task_iter_start(struct cgroup_subsys_state *css, unsigned int flags,
4192 struct css_task_iter *it)
4194 /* no one should try to iterate before mounting cgroups */
4195 WARN_ON_ONCE(!use_task_css_set_links);
4197 memset(it, 0, sizeof(*it));
4199 spin_lock_irq(&css_set_lock);
4205 it->cset_pos = &css->cgroup->e_csets[css->ss->id];
4207 it->cset_pos = &css->cgroup->cset_links;
4209 it->cset_head = it->cset_pos;
4211 css_task_iter_advance(it);
4213 spin_unlock_irq(&css_set_lock);
4217 * css_task_iter_next - return the next task for the iterator
4218 * @it: the task iterator being iterated
4220 * The "next" function for task iteration. @it should have been
4221 * initialized via css_task_iter_start(). Returns NULL when the iteration
4224 struct task_struct *css_task_iter_next(struct css_task_iter *it)
4227 put_task_struct(it->cur_task);
4228 it->cur_task = NULL;
4231 spin_lock_irq(&css_set_lock);
4233 /* @it may be half-advanced by skips, finish advancing */
4234 if (it->flags & CSS_TASK_ITER_SKIPPED)
4235 css_task_iter_advance(it);
4238 it->cur_task = list_entry(it->task_pos, struct task_struct,
4240 get_task_struct(it->cur_task);
4241 css_task_iter_advance(it);
4244 spin_unlock_irq(&css_set_lock);
4246 return it->cur_task;
4250 * css_task_iter_end - finish task iteration
4251 * @it: the task iterator to finish
4253 * Finish task iteration started by css_task_iter_start().
4255 void css_task_iter_end(struct css_task_iter *it)
4258 spin_lock_irq(&css_set_lock);
4259 list_del(&it->iters_node);
4260 put_css_set_locked(it->cur_cset);
4261 spin_unlock_irq(&css_set_lock);
4265 put_css_set(it->cur_dcset);
4268 put_task_struct(it->cur_task);
4271 static void cgroup_procs_release(struct kernfs_open_file *of)
4274 css_task_iter_end(of->priv);
4279 static void *cgroup_procs_next(struct seq_file *s, void *v, loff_t *pos)
4281 struct kernfs_open_file *of = s->private;
4282 struct css_task_iter *it = of->priv;
4287 return css_task_iter_next(it);
4290 static void *__cgroup_procs_start(struct seq_file *s, loff_t *pos,
4291 unsigned int iter_flags)
4293 struct kernfs_open_file *of = s->private;
4294 struct cgroup *cgrp = seq_css(s)->cgroup;
4295 struct css_task_iter *it = of->priv;
4298 * When a seq_file is seeked, it's always traversed sequentially
4299 * from position 0, so we can simply keep iterating on !0 *pos.
4302 if (WARN_ON_ONCE((*pos)))
4303 return ERR_PTR(-EINVAL);
4305 it = kzalloc(sizeof(*it), GFP_KERNEL);
4307 return ERR_PTR(-ENOMEM);
4309 css_task_iter_start(&cgrp->self, iter_flags, it);
4310 } else if (!(*pos)) {
4311 css_task_iter_end(it);
4312 css_task_iter_start(&cgrp->self, iter_flags, it);
4314 return it->cur_task;
4316 return cgroup_procs_next(s, NULL, NULL);
4319 static void *cgroup_procs_start(struct seq_file *s, loff_t *pos)
4321 struct cgroup *cgrp = seq_css(s)->cgroup;
4324 * All processes of a threaded subtree belong to the domain cgroup
4325 * of the subtree. Only threads can be distributed across the
4326 * subtree. Reject reads on cgroup.procs in the subtree proper.
4327 * They're always empty anyway.
4329 if (cgroup_is_threaded(cgrp))
4330 return ERR_PTR(-EOPNOTSUPP);
4332 return __cgroup_procs_start(s, pos, CSS_TASK_ITER_PROCS |
4333 CSS_TASK_ITER_THREADED);
4336 static int cgroup_procs_show(struct seq_file *s, void *v)
4338 seq_printf(s, "%d\n", task_pid_vnr(v));
4342 static int cgroup_procs_write_permission(struct cgroup *src_cgrp,
4343 struct cgroup *dst_cgrp,
4344 struct super_block *sb)
4346 struct cgroup_namespace *ns = current->nsproxy->cgroup_ns;
4347 struct cgroup *com_cgrp = src_cgrp;
4348 struct inode *inode;
4351 lockdep_assert_held(&cgroup_mutex);
4353 /* find the common ancestor */
4354 while (!cgroup_is_descendant(dst_cgrp, com_cgrp))
4355 com_cgrp = cgroup_parent(com_cgrp);
4357 /* %current should be authorized to migrate to the common ancestor */
4358 inode = kernfs_get_inode(sb, com_cgrp->procs_file.kn);
4362 ret = inode_permission(inode, MAY_WRITE);
4368 * If namespaces are delegation boundaries, %current must be able
4369 * to see both source and destination cgroups from its namespace.
4371 if ((cgrp_dfl_root.flags & CGRP_ROOT_NS_DELEGATE) &&
4372 (!cgroup_is_descendant(src_cgrp, ns->root_cset->dfl_cgrp) ||
4373 !cgroup_is_descendant(dst_cgrp, ns->root_cset->dfl_cgrp)))
4379 static ssize_t cgroup_procs_write(struct kernfs_open_file *of,
4380 char *buf, size_t nbytes, loff_t off)
4382 struct cgroup *src_cgrp, *dst_cgrp;
4383 struct task_struct *task;
4386 dst_cgrp = cgroup_kn_lock_live(of->kn, false);
4390 task = cgroup_procs_write_start(buf, true);
4391 ret = PTR_ERR_OR_ZERO(task);
4395 /* find the source cgroup */
4396 spin_lock_irq(&css_set_lock);
4397 src_cgrp = task_cgroup_from_root(task, &cgrp_dfl_root);
4398 spin_unlock_irq(&css_set_lock);
4400 ret = cgroup_procs_write_permission(src_cgrp, dst_cgrp,
4401 of->file->f_path.dentry->d_sb);
4405 ret = cgroup_attach_task(dst_cgrp, task, true);
4408 cgroup_procs_write_finish(task);
4410 cgroup_kn_unlock(of->kn);
4412 return ret ?: nbytes;
4415 static void *cgroup_threads_start(struct seq_file *s, loff_t *pos)
4417 return __cgroup_procs_start(s, pos, 0);
4420 static ssize_t cgroup_threads_write(struct kernfs_open_file *of,
4421 char *buf, size_t nbytes, loff_t off)
4423 struct cgroup *src_cgrp, *dst_cgrp;
4424 struct task_struct *task;
4427 buf = strstrip(buf);
4429 dst_cgrp = cgroup_kn_lock_live(of->kn, false);
4433 task = cgroup_procs_write_start(buf, false);
4434 ret = PTR_ERR_OR_ZERO(task);
4438 /* find the source cgroup */
4439 spin_lock_irq(&css_set_lock);
4440 src_cgrp = task_cgroup_from_root(task, &cgrp_dfl_root);
4441 spin_unlock_irq(&css_set_lock);
4443 /* thread migrations follow the cgroup.procs delegation rule */
4444 ret = cgroup_procs_write_permission(src_cgrp, dst_cgrp,
4445 of->file->f_path.dentry->d_sb);
4449 /* and must be contained in the same domain */
4451 if (src_cgrp->dom_cgrp != dst_cgrp->dom_cgrp)
4454 ret = cgroup_attach_task(dst_cgrp, task, false);
4457 cgroup_procs_write_finish(task);
4459 cgroup_kn_unlock(of->kn);
4461 return ret ?: nbytes;
4464 /* cgroup core interface files for the default hierarchy */
4465 static struct cftype cgroup_base_files[] = {
4467 .name = "cgroup.type",
4468 .flags = CFTYPE_NOT_ON_ROOT,
4469 .seq_show = cgroup_type_show,
4470 .write = cgroup_type_write,
4473 .name = "cgroup.procs",
4474 .flags = CFTYPE_NS_DELEGATABLE,
4475 .file_offset = offsetof(struct cgroup, procs_file),
4476 .release = cgroup_procs_release,
4477 .seq_start = cgroup_procs_start,
4478 .seq_next = cgroup_procs_next,
4479 .seq_show = cgroup_procs_show,
4480 .write = cgroup_procs_write,
4483 .name = "cgroup.threads",
4484 .release = cgroup_procs_release,
4485 .seq_start = cgroup_threads_start,
4486 .seq_next = cgroup_procs_next,
4487 .seq_show = cgroup_procs_show,
4488 .write = cgroup_threads_write,
4491 .name = "cgroup.controllers",
4492 .seq_show = cgroup_controllers_show,
4495 .name = "cgroup.subtree_control",
4496 .flags = CFTYPE_NS_DELEGATABLE,
4497 .seq_show = cgroup_subtree_control_show,
4498 .write = cgroup_subtree_control_write,
4501 .name = "cgroup.events",
4502 .flags = CFTYPE_NOT_ON_ROOT,
4503 .file_offset = offsetof(struct cgroup, events_file),
4504 .seq_show = cgroup_events_show,
4507 .name = "cgroup.max.descendants",
4508 .seq_show = cgroup_max_descendants_show,
4509 .write = cgroup_max_descendants_write,
4512 .name = "cgroup.max.depth",
4513 .seq_show = cgroup_max_depth_show,
4514 .write = cgroup_max_depth_write,
4517 .name = "cgroup.stat",
4518 .seq_show = cgroup_stat_show,
4524 * css destruction is four-stage process.
4526 * 1. Destruction starts. Killing of the percpu_ref is initiated.
4527 * Implemented in kill_css().
4529 * 2. When the percpu_ref is confirmed to be visible as killed on all CPUs
4530 * and thus css_tryget_online() is guaranteed to fail, the css can be
4531 * offlined by invoking offline_css(). After offlining, the base ref is
4532 * put. Implemented in css_killed_work_fn().
4534 * 3. When the percpu_ref reaches zero, the only possible remaining
4535 * accessors are inside RCU read sections. css_release() schedules the
4538 * 4. After the grace period, the css can be freed. Implemented in
4539 * css_free_work_fn().
4541 * It is actually hairier because both step 2 and 4 require process context
4542 * and thus involve punting to css->destroy_work adding two additional
4543 * steps to the already complex sequence.
4545 static void css_free_work_fn(struct work_struct *work)
4547 struct cgroup_subsys_state *css =
4548 container_of(work, struct cgroup_subsys_state, destroy_work);
4549 struct cgroup_subsys *ss = css->ss;
4550 struct cgroup *cgrp = css->cgroup;
4552 percpu_ref_exit(&css->refcnt);
4556 struct cgroup_subsys_state *parent = css->parent;
4560 cgroup_idr_remove(&ss->css_idr, id);
4566 /* cgroup free path */
4567 atomic_dec(&cgrp->root->nr_cgrps);
4568 cgroup1_pidlist_destroy_all(cgrp);
4569 cancel_work_sync(&cgrp->release_agent_work);
4571 if (cgroup_parent(cgrp)) {
4573 * We get a ref to the parent, and put the ref when
4574 * this cgroup is being freed, so it's guaranteed
4575 * that the parent won't be destroyed before its
4578 cgroup_put(cgroup_parent(cgrp));
4579 kernfs_put(cgrp->kn);
4583 * This is root cgroup's refcnt reaching zero,
4584 * which indicates that the root should be
4587 cgroup_destroy_root(cgrp->root);
4592 static void css_free_rcu_fn(struct rcu_head *rcu_head)
4594 struct cgroup_subsys_state *css =
4595 container_of(rcu_head, struct cgroup_subsys_state, rcu_head);
4597 INIT_WORK(&css->destroy_work, css_free_work_fn);
4598 queue_work(cgroup_destroy_wq, &css->destroy_work);
4601 static void css_release_work_fn(struct work_struct *work)
4603 struct cgroup_subsys_state *css =
4604 container_of(work, struct cgroup_subsys_state, destroy_work);
4605 struct cgroup_subsys *ss = css->ss;
4606 struct cgroup *cgrp = css->cgroup;
4608 mutex_lock(&cgroup_mutex);
4610 css->flags |= CSS_RELEASED;
4611 list_del_rcu(&css->sibling);
4614 /* css release path */
4615 cgroup_idr_replace(&ss->css_idr, NULL, css->id);
4616 if (ss->css_released)
4617 ss->css_released(css);
4619 struct cgroup *tcgrp;
4621 /* cgroup release path */
4622 trace_cgroup_release(cgrp);
4624 spin_lock_irq(&css_set_lock);
4625 for (tcgrp = cgroup_parent(cgrp); tcgrp;
4626 tcgrp = cgroup_parent(tcgrp))
4627 tcgrp->nr_dying_descendants--;
4628 spin_unlock_irq(&css_set_lock);
4630 cgroup_idr_remove(&cgrp->root->cgroup_idr, cgrp->id);
4634 * There are two control paths which try to determine
4635 * cgroup from dentry without going through kernfs -
4636 * cgroupstats_build() and css_tryget_online_from_dir().
4637 * Those are supported by RCU protecting clearing of
4638 * cgrp->kn->priv backpointer.
4641 RCU_INIT_POINTER(*(void __rcu __force **)&cgrp->kn->priv,
4644 cgroup_bpf_put(cgrp);
4647 mutex_unlock(&cgroup_mutex);
4649 call_rcu(&css->rcu_head, css_free_rcu_fn);
4652 static void css_release(struct percpu_ref *ref)
4654 struct cgroup_subsys_state *css =
4655 container_of(ref, struct cgroup_subsys_state, refcnt);
4657 INIT_WORK(&css->destroy_work, css_release_work_fn);
4658 queue_work(cgroup_destroy_wq, &css->destroy_work);
4661 static void init_and_link_css(struct cgroup_subsys_state *css,
4662 struct cgroup_subsys *ss, struct cgroup *cgrp)
4664 lockdep_assert_held(&cgroup_mutex);
4666 cgroup_get_live(cgrp);
4668 memset(css, 0, sizeof(*css));
4672 INIT_LIST_HEAD(&css->sibling);
4673 INIT_LIST_HEAD(&css->children);
4674 css->serial_nr = css_serial_nr_next++;
4675 atomic_set(&css->online_cnt, 0);
4677 if (cgroup_parent(cgrp)) {
4678 css->parent = cgroup_css(cgroup_parent(cgrp), ss);
4679 css_get(css->parent);
4682 BUG_ON(cgroup_css(cgrp, ss));
4685 /* invoke ->css_online() on a new CSS and mark it online if successful */
4686 static int online_css(struct cgroup_subsys_state *css)
4688 struct cgroup_subsys *ss = css->ss;
4691 lockdep_assert_held(&cgroup_mutex);
4694 ret = ss->css_online(css);
4696 css->flags |= CSS_ONLINE;
4697 rcu_assign_pointer(css->cgroup->subsys[ss->id], css);
4699 atomic_inc(&css->online_cnt);
4701 atomic_inc(&css->parent->online_cnt);
4706 /* if the CSS is online, invoke ->css_offline() on it and mark it offline */
4707 static void offline_css(struct cgroup_subsys_state *css)
4709 struct cgroup_subsys *ss = css->ss;
4711 lockdep_assert_held(&cgroup_mutex);
4713 if (!(css->flags & CSS_ONLINE))
4716 if (ss->css_offline)
4717 ss->css_offline(css);
4719 css->flags &= ~CSS_ONLINE;
4720 RCU_INIT_POINTER(css->cgroup->subsys[ss->id], NULL);
4722 wake_up_all(&css->cgroup->offline_waitq);
4726 * css_create - create a cgroup_subsys_state
4727 * @cgrp: the cgroup new css will be associated with
4728 * @ss: the subsys of new css
4730 * Create a new css associated with @cgrp - @ss pair. On success, the new
4731 * css is online and installed in @cgrp. This function doesn't create the
4732 * interface files. Returns 0 on success, -errno on failure.
4734 static struct cgroup_subsys_state *css_create(struct cgroup *cgrp,
4735 struct cgroup_subsys *ss)
4737 struct cgroup *parent = cgroup_parent(cgrp);
4738 struct cgroup_subsys_state *parent_css = cgroup_css(parent, ss);
4739 struct cgroup_subsys_state *css;
4742 lockdep_assert_held(&cgroup_mutex);
4744 css = ss->css_alloc(parent_css);
4746 css = ERR_PTR(-ENOMEM);
4750 init_and_link_css(css, ss, cgrp);
4752 err = percpu_ref_init(&css->refcnt, css_release, 0, GFP_KERNEL);
4756 err = cgroup_idr_alloc(&ss->css_idr, NULL, 2, 0, GFP_KERNEL);
4761 /* @css is ready to be brought online now, make it visible */
4762 list_add_tail_rcu(&css->sibling, &parent_css->children);
4763 cgroup_idr_replace(&ss->css_idr, css, css->id);
4765 err = online_css(css);
4769 if (ss->broken_hierarchy && !ss->warned_broken_hierarchy &&
4770 cgroup_parent(parent)) {
4771 pr_warn("%s (%d) created nested cgroup for controller \"%s\" which has incomplete hierarchy support. Nested cgroups may change behavior in the future.\n",
4772 current->comm, current->pid, ss->name);
4773 if (!strcmp(ss->name, "memory"))
4774 pr_warn("\"memory\" requires setting use_hierarchy to 1 on the root\n");
4775 ss->warned_broken_hierarchy = true;
4781 list_del_rcu(&css->sibling);
4783 call_rcu(&css->rcu_head, css_free_rcu_fn);
4784 return ERR_PTR(err);
4788 * The returned cgroup is fully initialized including its control mask, but
4789 * it isn't associated with its kernfs_node and doesn't have the control
4792 static struct cgroup *cgroup_create(struct cgroup *parent)
4794 struct cgroup_root *root = parent->root;
4795 struct cgroup *cgrp, *tcgrp;
4796 int level = parent->level + 1;
4799 /* allocate the cgroup and its ID, 0 is reserved for the root */
4800 cgrp = kzalloc(sizeof(*cgrp) +
4801 sizeof(cgrp->ancestor_ids[0]) * (level + 1), GFP_KERNEL);
4803 return ERR_PTR(-ENOMEM);
4805 ret = percpu_ref_init(&cgrp->self.refcnt, css_release, 0, GFP_KERNEL);
4810 * Temporarily set the pointer to NULL, so idr_find() won't return
4811 * a half-baked cgroup.
4813 cgrp->id = cgroup_idr_alloc(&root->cgroup_idr, NULL, 2, 0, GFP_KERNEL);
4816 goto out_cancel_ref;
4819 init_cgroup_housekeeping(cgrp);
4821 cgrp->self.parent = &parent->self;
4823 cgrp->level = level;
4825 spin_lock_irq(&css_set_lock);
4826 for (tcgrp = cgrp; tcgrp; tcgrp = cgroup_parent(tcgrp)) {
4827 cgrp->ancestor_ids[tcgrp->level] = tcgrp->id;
4830 tcgrp->nr_descendants++;
4832 spin_unlock_irq(&css_set_lock);
4834 if (notify_on_release(parent))
4835 set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
4837 if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &parent->flags))
4838 set_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags);
4840 cgrp->self.serial_nr = css_serial_nr_next++;
4842 /* allocation complete, commit to creation */
4843 list_add_tail_rcu(&cgrp->self.sibling, &cgroup_parent(cgrp)->self.children);
4844 atomic_inc(&root->nr_cgrps);
4845 cgroup_get_live(parent);
4848 * @cgrp is now fully operational. If something fails after this
4849 * point, it'll be released via the normal destruction path.
4851 cgroup_idr_replace(&root->cgroup_idr, cgrp, cgrp->id);
4854 * On the default hierarchy, a child doesn't automatically inherit
4855 * subtree_control from the parent. Each is configured manually.
4857 if (!cgroup_on_dfl(cgrp))
4858 cgrp->subtree_control = cgroup_control(cgrp);
4861 cgroup_bpf_inherit(cgrp, parent);
4863 cgroup_propagate_control(cgrp);
4868 percpu_ref_exit(&cgrp->self.refcnt);
4871 return ERR_PTR(ret);
4874 static bool cgroup_check_hierarchy_limits(struct cgroup *parent)
4876 struct cgroup *cgroup;
4880 lockdep_assert_held(&cgroup_mutex);
4882 for (cgroup = parent; cgroup; cgroup = cgroup_parent(cgroup)) {
4883 if (cgroup->nr_descendants >= cgroup->max_descendants)
4886 if (level > cgroup->max_depth)
4897 int cgroup_mkdir(struct kernfs_node *parent_kn, const char *name, umode_t mode)
4899 struct cgroup *parent, *cgrp;
4900 struct kernfs_node *kn;
4903 /* do not accept '\n' to prevent making /proc/<pid>/cgroup unparsable */
4904 if (strchr(name, '\n'))
4907 parent = cgroup_kn_lock_live(parent_kn, false);
4911 if (!cgroup_check_hierarchy_limits(parent)) {
4916 cgrp = cgroup_create(parent);
4918 ret = PTR_ERR(cgrp);
4922 /* create the directory */
4923 kn = kernfs_create_dir(parent->kn, name, mode, cgrp);
4931 * This extra ref will be put in cgroup_free_fn() and guarantees
4932 * that @cgrp->kn is always accessible.
4936 ret = cgroup_kn_set_ugid(kn);
4940 ret = css_populate_dir(&cgrp->self);
4944 ret = cgroup_apply_control_enable(cgrp);
4948 trace_cgroup_mkdir(cgrp);
4950 /* let's create and online css's */
4951 kernfs_activate(kn);
4957 cgroup_destroy_locked(cgrp);
4959 cgroup_kn_unlock(parent_kn);
4964 * This is called when the refcnt of a css is confirmed to be killed.
4965 * css_tryget_online() is now guaranteed to fail. Tell the subsystem to
4966 * initate destruction and put the css ref from kill_css().
4968 static void css_killed_work_fn(struct work_struct *work)
4970 struct cgroup_subsys_state *css =
4971 container_of(work, struct cgroup_subsys_state, destroy_work);
4973 mutex_lock(&cgroup_mutex);
4978 /* @css can't go away while we're holding cgroup_mutex */
4980 } while (css && atomic_dec_and_test(&css->online_cnt));
4982 mutex_unlock(&cgroup_mutex);
4985 /* css kill confirmation processing requires process context, bounce */
4986 static void css_killed_ref_fn(struct percpu_ref *ref)
4988 struct cgroup_subsys_state *css =
4989 container_of(ref, struct cgroup_subsys_state, refcnt);
4991 if (atomic_dec_and_test(&css->online_cnt)) {
4992 INIT_WORK(&css->destroy_work, css_killed_work_fn);
4993 queue_work(cgroup_destroy_wq, &css->destroy_work);
4998 * kill_css - destroy a css
4999 * @css: css to destroy
5001 * This function initiates destruction of @css by removing cgroup interface
5002 * files and putting its base reference. ->css_offline() will be invoked
5003 * asynchronously once css_tryget_online() is guaranteed to fail and when
5004 * the reference count reaches zero, @css will be released.
5006 static void kill_css(struct cgroup_subsys_state *css)
5008 lockdep_assert_held(&cgroup_mutex);
5010 if (css->flags & CSS_DYING)
5013 css->flags |= CSS_DYING;
5016 * This must happen before css is disassociated with its cgroup.
5017 * See seq_css() for details.
5022 * Killing would put the base ref, but we need to keep it alive
5023 * until after ->css_offline().
5028 * cgroup core guarantees that, by the time ->css_offline() is
5029 * invoked, no new css reference will be given out via
5030 * css_tryget_online(). We can't simply call percpu_ref_kill() and
5031 * proceed to offlining css's because percpu_ref_kill() doesn't
5032 * guarantee that the ref is seen as killed on all CPUs on return.
5034 * Use percpu_ref_kill_and_confirm() to get notifications as each
5035 * css is confirmed to be seen as killed on all CPUs.
5037 percpu_ref_kill_and_confirm(&css->refcnt, css_killed_ref_fn);
5041 * cgroup_destroy_locked - the first stage of cgroup destruction
5042 * @cgrp: cgroup to be destroyed
5044 * css's make use of percpu refcnts whose killing latency shouldn't be
5045 * exposed to userland and are RCU protected. Also, cgroup core needs to
5046 * guarantee that css_tryget_online() won't succeed by the time
5047 * ->css_offline() is invoked. To satisfy all the requirements,
5048 * destruction is implemented in the following two steps.
5050 * s1. Verify @cgrp can be destroyed and mark it dying. Remove all
5051 * userland visible parts and start killing the percpu refcnts of
5052 * css's. Set up so that the next stage will be kicked off once all
5053 * the percpu refcnts are confirmed to be killed.
5055 * s2. Invoke ->css_offline(), mark the cgroup dead and proceed with the
5056 * rest of destruction. Once all cgroup references are gone, the
5057 * cgroup is RCU-freed.
5059 * This function implements s1. After this step, @cgrp is gone as far as
5060 * the userland is concerned and a new cgroup with the same name may be
5061 * created. As cgroup doesn't care about the names internally, this
5062 * doesn't cause any problem.
5064 static int cgroup_destroy_locked(struct cgroup *cgrp)
5065 __releases(&cgroup_mutex) __acquires(&cgroup_mutex)
5067 struct cgroup *tcgrp, *parent = cgroup_parent(cgrp);
5068 struct cgroup_subsys_state *css;
5069 struct cgrp_cset_link *link;
5072 lockdep_assert_held(&cgroup_mutex);
5075 * Only migration can raise populated from zero and we're already
5076 * holding cgroup_mutex.
5078 if (cgroup_is_populated(cgrp))
5082 * Make sure there's no live children. We can't test emptiness of
5083 * ->self.children as dead children linger on it while being
5084 * drained; otherwise, "rmdir parent/child parent" may fail.
5086 if (css_has_online_children(&cgrp->self))
5090 * Mark @cgrp and the associated csets dead. The former prevents
5091 * further task migration and child creation by disabling
5092 * cgroup_lock_live_group(). The latter makes the csets ignored by
5093 * the migration path.
5095 cgrp->self.flags &= ~CSS_ONLINE;
5097 spin_lock_irq(&css_set_lock);
5098 list_for_each_entry(link, &cgrp->cset_links, cset_link)
5099 link->cset->dead = true;
5100 spin_unlock_irq(&css_set_lock);
5102 /* initiate massacre of all css's */
5103 for_each_css(css, ssid, cgrp)
5107 * Remove @cgrp directory along with the base files. @cgrp has an
5108 * extra ref on its kn.
5110 kernfs_remove(cgrp->kn);
5112 if (parent && cgroup_is_threaded(cgrp))
5113 parent->nr_threaded_children--;
5115 spin_lock_irq(&css_set_lock);
5116 for (tcgrp = cgroup_parent(cgrp); tcgrp; tcgrp = cgroup_parent(tcgrp)) {
5117 tcgrp->nr_descendants--;
5118 tcgrp->nr_dying_descendants++;
5120 spin_unlock_irq(&css_set_lock);
5122 cgroup1_check_for_release(parent);
5124 /* put the base reference */
5125 percpu_ref_kill(&cgrp->self.refcnt);
5130 int cgroup_rmdir(struct kernfs_node *kn)
5132 struct cgroup *cgrp;
5135 cgrp = cgroup_kn_lock_live(kn, false);
5139 ret = cgroup_destroy_locked(cgrp);
5142 trace_cgroup_rmdir(cgrp);
5144 cgroup_kn_unlock(kn);
5148 static struct kernfs_syscall_ops cgroup_kf_syscall_ops = {
5149 .show_options = cgroup_show_options,
5150 .remount_fs = cgroup_remount,
5151 .mkdir = cgroup_mkdir,
5152 .rmdir = cgroup_rmdir,
5153 .show_path = cgroup_show_path,
5156 static void __init cgroup_init_subsys(struct cgroup_subsys *ss, bool early)
5158 struct cgroup_subsys_state *css;
5160 pr_debug("Initializing cgroup subsys %s\n", ss->name);
5162 mutex_lock(&cgroup_mutex);
5164 idr_init(&ss->css_idr);
5165 INIT_LIST_HEAD(&ss->cfts);
5167 /* Create the root cgroup state for this subsystem */
5168 ss->root = &cgrp_dfl_root;
5169 css = ss->css_alloc(cgroup_css(&cgrp_dfl_root.cgrp, ss));
5170 /* We don't handle early failures gracefully */
5171 BUG_ON(IS_ERR(css));
5172 init_and_link_css(css, ss, &cgrp_dfl_root.cgrp);
5175 * Root csses are never destroyed and we can't initialize
5176 * percpu_ref during early init. Disable refcnting.
5178 css->flags |= CSS_NO_REF;
5181 /* allocation can't be done safely during early init */
5184 css->id = cgroup_idr_alloc(&ss->css_idr, css, 1, 2, GFP_KERNEL);
5185 BUG_ON(css->id < 0);
5188 /* Update the init_css_set to contain a subsys
5189 * pointer to this state - since the subsystem is
5190 * newly registered, all tasks and hence the
5191 * init_css_set is in the subsystem's root cgroup. */
5192 init_css_set.subsys[ss->id] = css;
5194 have_fork_callback |= (bool)ss->fork << ss->id;
5195 have_exit_callback |= (bool)ss->exit << ss->id;
5196 have_release_callback |= (bool)ss->release << ss->id;
5197 have_canfork_callback |= (bool)ss->can_fork << ss->id;
5199 /* At system boot, before all subsystems have been
5200 * registered, no tasks have been forked, so we don't
5201 * need to invoke fork callbacks here. */
5202 BUG_ON(!list_empty(&init_task.tasks));
5204 BUG_ON(online_css(css));
5206 mutex_unlock(&cgroup_mutex);
5210 * cgroup_init_early - cgroup initialization at system boot
5212 * Initialize cgroups at system boot, and initialize any
5213 * subsystems that request early init.
5215 int __init cgroup_init_early(void)
5217 static struct cgroup_sb_opts __initdata opts;
5218 struct cgroup_subsys *ss;
5221 init_cgroup_root(&cgrp_dfl_root, &opts);
5222 cgrp_dfl_root.cgrp.self.flags |= CSS_NO_REF;
5224 RCU_INIT_POINTER(init_task.cgroups, &init_css_set);
5226 for_each_subsys(ss, i) {
5227 WARN(!ss->css_alloc || !ss->css_free || ss->name || ss->id,
5228 "invalid cgroup_subsys %d:%s css_alloc=%p css_free=%p id:name=%d:%s\n",
5229 i, cgroup_subsys_name[i], ss->css_alloc, ss->css_free,
5231 WARN(strlen(cgroup_subsys_name[i]) > MAX_CGROUP_TYPE_NAMELEN,
5232 "cgroup_subsys_name %s too long\n", cgroup_subsys_name[i]);
5235 ss->name = cgroup_subsys_name[i];
5236 if (!ss->legacy_name)
5237 ss->legacy_name = cgroup_subsys_name[i];
5240 cgroup_init_subsys(ss, true);
5246 * cgroup_init - cgroup initialization
5248 * Register cgroup filesystem and /proc file, and initialize
5249 * any subsystems that didn't request early init.
5251 int __init cgroup_init(void)
5253 struct cgroup_subsys *ss;
5256 BUILD_BUG_ON(CGROUP_SUBSYS_COUNT > 16);
5257 BUG_ON(percpu_init_rwsem(&cgroup_threadgroup_rwsem));
5258 BUG_ON(cgroup_init_cftypes(NULL, cgroup_base_files));
5259 BUG_ON(cgroup_init_cftypes(NULL, cgroup1_base_files));
5262 * The latency of the synchronize_sched() is too high for cgroups,
5263 * avoid it at the cost of forcing all readers into the slow path.
5265 rcu_sync_enter_start(&cgroup_threadgroup_rwsem.rss);
5267 get_user_ns(init_cgroup_ns.user_ns);
5269 mutex_lock(&cgroup_mutex);
5272 * Add init_css_set to the hash table so that dfl_root can link to
5275 hash_add(css_set_table, &init_css_set.hlist,
5276 css_set_hash(init_css_set.subsys));
5278 BUG_ON(cgroup_setup_root(&cgrp_dfl_root, 0, 0));
5280 mutex_unlock(&cgroup_mutex);
5282 for_each_subsys(ss, ssid) {
5283 if (ss->early_init) {
5284 struct cgroup_subsys_state *css =
5285 init_css_set.subsys[ss->id];
5287 css->id = cgroup_idr_alloc(&ss->css_idr, css, 1, 2,
5289 BUG_ON(css->id < 0);
5291 cgroup_init_subsys(ss, false);
5294 list_add_tail(&init_css_set.e_cset_node[ssid],
5295 &cgrp_dfl_root.cgrp.e_csets[ssid]);
5298 * Setting dfl_root subsys_mask needs to consider the
5299 * disabled flag and cftype registration needs kmalloc,
5300 * both of which aren't available during early_init.
5302 if (!cgroup_ssid_enabled(ssid))
5305 if (cgroup1_ssid_disabled(ssid))
5306 printk(KERN_INFO "Disabling %s control group subsystem in v1 mounts\n",
5309 cgrp_dfl_root.subsys_mask |= 1 << ss->id;
5311 /* implicit controllers must be threaded too */
5312 WARN_ON(ss->implicit_on_dfl && !ss->threaded);
5314 if (ss->implicit_on_dfl)
5315 cgrp_dfl_implicit_ss_mask |= 1 << ss->id;
5316 else if (!ss->dfl_cftypes)
5317 cgrp_dfl_inhibit_ss_mask |= 1 << ss->id;
5320 cgrp_dfl_threaded_ss_mask |= 1 << ss->id;
5322 if (ss->dfl_cftypes == ss->legacy_cftypes) {
5323 WARN_ON(cgroup_add_cftypes(ss, ss->dfl_cftypes));
5325 WARN_ON(cgroup_add_dfl_cftypes(ss, ss->dfl_cftypes));
5326 WARN_ON(cgroup_add_legacy_cftypes(ss, ss->legacy_cftypes));
5330 ss->bind(init_css_set.subsys[ssid]);
5332 mutex_lock(&cgroup_mutex);
5333 css_populate_dir(init_css_set.subsys[ssid]);
5334 mutex_unlock(&cgroup_mutex);
5337 /* init_css_set.subsys[] has been updated, re-hash */
5338 hash_del(&init_css_set.hlist);
5339 hash_add(css_set_table, &init_css_set.hlist,
5340 css_set_hash(init_css_set.subsys));
5342 WARN_ON(sysfs_create_mount_point(fs_kobj, "cgroup"));
5343 WARN_ON(register_filesystem(&cgroup_fs_type));
5344 WARN_ON(register_filesystem(&cgroup2_fs_type));
5345 WARN_ON(!proc_create("cgroups", 0, NULL, &proc_cgroupstats_operations));
5350 static int __init cgroup_wq_init(void)
5353 * There isn't much point in executing destruction path in
5354 * parallel. Good chunk is serialized with cgroup_mutex anyway.
5355 * Use 1 for @max_active.
5357 * We would prefer to do this in cgroup_init() above, but that
5358 * is called before init_workqueues(): so leave this until after.
5360 cgroup_destroy_wq = alloc_workqueue("cgroup_destroy", 0, 1);
5361 BUG_ON(!cgroup_destroy_wq);
5364 core_initcall(cgroup_wq_init);
5366 void cgroup_path_from_kernfs_id(const union kernfs_node_id *id,
5367 char *buf, size_t buflen)
5369 struct kernfs_node *kn;
5371 kn = kernfs_get_node_by_id(cgrp_dfl_root.kf_root, id);
5374 kernfs_path(kn, buf, buflen);
5379 * proc_cgroup_show()
5380 * - Print task's cgroup paths into seq_file, one line for each hierarchy
5381 * - Used for /proc/<pid>/cgroup.
5383 int proc_cgroup_show(struct seq_file *m, struct pid_namespace *ns,
5384 struct pid *pid, struct task_struct *tsk)
5388 struct cgroup_root *root;
5391 buf = kmalloc(PATH_MAX, GFP_KERNEL);
5395 mutex_lock(&cgroup_mutex);
5396 spin_lock_irq(&css_set_lock);
5398 for_each_root(root) {
5399 struct cgroup_subsys *ss;
5400 struct cgroup *cgrp;
5401 int ssid, count = 0;
5403 if (root == &cgrp_dfl_root && !cgrp_dfl_visible)
5406 seq_printf(m, "%d:", root->hierarchy_id);
5407 if (root != &cgrp_dfl_root)
5408 for_each_subsys(ss, ssid)
5409 if (root->subsys_mask & (1 << ssid))
5410 seq_printf(m, "%s%s", count++ ? "," : "",
5412 if (strlen(root->name))
5413 seq_printf(m, "%sname=%s", count ? "," : "",
5417 cgrp = task_cgroup_from_root(tsk, root);
5420 * On traditional hierarchies, all zombie tasks show up as
5421 * belonging to the root cgroup. On the default hierarchy,
5422 * while a zombie doesn't show up in "cgroup.procs" and
5423 * thus can't be migrated, its /proc/PID/cgroup keeps
5424 * reporting the cgroup it belonged to before exiting. If
5425 * the cgroup is removed before the zombie is reaped,
5426 * " (deleted)" is appended to the cgroup path.
5428 if (cgroup_on_dfl(cgrp) || !(tsk->flags & PF_EXITING)) {
5429 retval = cgroup_path_ns_locked(cgrp, buf, PATH_MAX,
5430 current->nsproxy->cgroup_ns);
5431 if (retval >= PATH_MAX)
5432 retval = -ENAMETOOLONG;
5441 if (cgroup_on_dfl(cgrp) && cgroup_is_dead(cgrp))
5442 seq_puts(m, " (deleted)\n");
5449 spin_unlock_irq(&css_set_lock);
5450 mutex_unlock(&cgroup_mutex);
5457 * cgroup_fork - initialize cgroup related fields during copy_process()
5458 * @child: pointer to task_struct of forking parent process.
5460 * A task is associated with the init_css_set until cgroup_post_fork()
5461 * attaches it to the parent's css_set. Empty cg_list indicates that
5462 * @child isn't holding reference to its css_set.
5464 void cgroup_fork(struct task_struct *child)
5466 RCU_INIT_POINTER(child->cgroups, &init_css_set);
5467 INIT_LIST_HEAD(&child->cg_list);
5471 * cgroup_can_fork - called on a new task before the process is exposed
5472 * @child: the task in question.
5474 * This calls the subsystem can_fork() callbacks. If the can_fork() callback
5475 * returns an error, the fork aborts with that error code. This allows for
5476 * a cgroup subsystem to conditionally allow or deny new forks.
5478 int cgroup_can_fork(struct task_struct *child)
5480 struct cgroup_subsys *ss;
5483 do_each_subsys_mask(ss, i, have_canfork_callback) {
5484 ret = ss->can_fork(child);
5487 } while_each_subsys_mask();
5492 for_each_subsys(ss, j) {
5495 if (ss->cancel_fork)
5496 ss->cancel_fork(child);
5503 * cgroup_cancel_fork - called if a fork failed after cgroup_can_fork()
5504 * @child: the task in question
5506 * This calls the cancel_fork() callbacks if a fork failed *after*
5507 * cgroup_can_fork() succeded.
5509 void cgroup_cancel_fork(struct task_struct *child)
5511 struct cgroup_subsys *ss;
5514 for_each_subsys(ss, i)
5515 if (ss->cancel_fork)
5516 ss->cancel_fork(child);
5520 * cgroup_post_fork - called on a new task after adding it to the task list
5521 * @child: the task in question
5523 * Adds the task to the list running through its css_set if necessary and
5524 * call the subsystem fork() callbacks. Has to be after the task is
5525 * visible on the task list in case we race with the first call to
5526 * cgroup_task_iter_start() - to guarantee that the new task ends up on its
5529 void cgroup_post_fork(struct task_struct *child)
5531 struct cgroup_subsys *ss;
5535 * This may race against cgroup_enable_task_cg_lists(). As that
5536 * function sets use_task_css_set_links before grabbing
5537 * tasklist_lock and we just went through tasklist_lock to add
5538 * @child, it's guaranteed that either we see the set
5539 * use_task_css_set_links or cgroup_enable_task_cg_lists() sees
5540 * @child during its iteration.
5542 * If we won the race, @child is associated with %current's
5543 * css_set. Grabbing css_set_lock guarantees both that the
5544 * association is stable, and, on completion of the parent's
5545 * migration, @child is visible in the source of migration or
5546 * already in the destination cgroup. This guarantee is necessary
5547 * when implementing operations which need to migrate all tasks of
5548 * a cgroup to another.
5550 * Note that if we lose to cgroup_enable_task_cg_lists(), @child
5551 * will remain in init_css_set. This is safe because all tasks are
5552 * in the init_css_set before cg_links is enabled and there's no
5553 * operation which transfers all tasks out of init_css_set.
5555 if (use_task_css_set_links) {
5556 struct css_set *cset;
5558 spin_lock_irq(&css_set_lock);
5559 cset = task_css_set(current);
5560 if (list_empty(&child->cg_list)) {
5563 css_set_move_task(child, NULL, cset, false);
5565 spin_unlock_irq(&css_set_lock);
5569 * Call ss->fork(). This must happen after @child is linked on
5570 * css_set; otherwise, @child might change state between ->fork()
5571 * and addition to css_set.
5573 do_each_subsys_mask(ss, i, have_fork_callback) {
5575 } while_each_subsys_mask();
5579 * cgroup_exit - detach cgroup from exiting task
5580 * @tsk: pointer to task_struct of exiting process
5582 * Description: Detach cgroup from @tsk and release it.
5584 * Note that cgroups marked notify_on_release force every task in
5585 * them to take the global cgroup_mutex mutex when exiting.
5586 * This could impact scaling on very large systems. Be reluctant to
5587 * use notify_on_release cgroups where very high task exit scaling
5588 * is required on large systems.
5590 * We set the exiting tasks cgroup to the root cgroup (top_cgroup). We
5591 * call cgroup_exit() while the task is still competent to handle
5592 * notify_on_release(), then leave the task attached to the root cgroup in
5593 * each hierarchy for the remainder of its exit. No need to bother with
5594 * init_css_set refcnting. init_css_set never goes away and we can't race
5595 * with migration path - PF_EXITING is visible to migration path.
5597 void cgroup_exit(struct task_struct *tsk)
5599 struct cgroup_subsys *ss;
5600 struct css_set *cset;
5604 * Unlink from @tsk from its css_set. As migration path can't race
5605 * with us, we can check css_set and cg_list without synchronization.
5607 cset = task_css_set(tsk);
5609 if (!list_empty(&tsk->cg_list)) {
5610 spin_lock_irq(&css_set_lock);
5611 css_set_move_task(tsk, cset, NULL, false);
5612 list_add_tail(&tsk->cg_list, &cset->dying_tasks);
5614 spin_unlock_irq(&css_set_lock);
5619 /* see cgroup_post_fork() for details */
5620 do_each_subsys_mask(ss, i, have_exit_callback) {
5622 } while_each_subsys_mask();
5625 void cgroup_release(struct task_struct *task)
5627 struct cgroup_subsys *ss;
5630 do_each_subsys_mask(ss, ssid, have_release_callback) {
5632 } while_each_subsys_mask();
5634 if (use_task_css_set_links) {
5635 spin_lock_irq(&css_set_lock);
5636 css_set_skip_task_iters(task_css_set(task), task);
5637 list_del_init(&task->cg_list);
5638 spin_unlock_irq(&css_set_lock);
5642 void cgroup_free(struct task_struct *task)
5644 struct css_set *cset = task_css_set(task);
5648 static int __init cgroup_disable(char *str)
5650 struct cgroup_subsys *ss;
5654 while ((token = strsep(&str, ",")) != NULL) {
5658 for_each_subsys(ss, i) {
5659 if (strcmp(token, ss->name) &&
5660 strcmp(token, ss->legacy_name))
5663 static_branch_disable(cgroup_subsys_enabled_key[i]);
5664 pr_info("Disabling %s control group subsystem\n",
5670 __setup("cgroup_disable=", cgroup_disable);
5673 * css_tryget_online_from_dir - get corresponding css from a cgroup dentry
5674 * @dentry: directory dentry of interest
5675 * @ss: subsystem of interest
5677 * If @dentry is a directory for a cgroup which has @ss enabled on it, try
5678 * to get the corresponding css and return it. If such css doesn't exist
5679 * or can't be pinned, an ERR_PTR value is returned.
5681 struct cgroup_subsys_state *css_tryget_online_from_dir(struct dentry *dentry,
5682 struct cgroup_subsys *ss)
5684 struct kernfs_node *kn = kernfs_node_from_dentry(dentry);
5685 struct file_system_type *s_type = dentry->d_sb->s_type;
5686 struct cgroup_subsys_state *css = NULL;
5687 struct cgroup *cgrp;
5689 /* is @dentry a cgroup dir? */
5690 if ((s_type != &cgroup_fs_type && s_type != &cgroup2_fs_type) ||
5691 !kn || kernfs_type(kn) != KERNFS_DIR)
5692 return ERR_PTR(-EBADF);
5697 * This path doesn't originate from kernfs and @kn could already
5698 * have been or be removed at any point. @kn->priv is RCU
5699 * protected for this access. See css_release_work_fn() for details.
5701 cgrp = rcu_dereference(*(void __rcu __force **)&kn->priv);
5703 css = cgroup_css(cgrp, ss);
5705 if (!css || !css_tryget_online(css))
5706 css = ERR_PTR(-ENOENT);
5713 * css_from_id - lookup css by id
5714 * @id: the cgroup id
5715 * @ss: cgroup subsys to be looked into
5717 * Returns the css if there's valid one with @id, otherwise returns NULL.
5718 * Should be called under rcu_read_lock().
5720 struct cgroup_subsys_state *css_from_id(int id, struct cgroup_subsys *ss)
5722 WARN_ON_ONCE(!rcu_read_lock_held());
5723 return idr_find(&ss->css_idr, id);
5727 * cgroup_get_from_path - lookup and get a cgroup from its default hierarchy path
5728 * @path: path on the default hierarchy
5730 * Find the cgroup at @path on the default hierarchy, increment its
5731 * reference count and return it. Returns pointer to the found cgroup on
5732 * success, ERR_PTR(-ENOENT) if @path doens't exist and ERR_PTR(-ENOTDIR)
5733 * if @path points to a non-directory.
5735 struct cgroup *cgroup_get_from_path(const char *path)
5737 struct kernfs_node *kn;
5738 struct cgroup *cgrp;
5740 mutex_lock(&cgroup_mutex);
5742 kn = kernfs_walk_and_get(cgrp_dfl_root.cgrp.kn, path);
5744 if (kernfs_type(kn) == KERNFS_DIR) {
5746 cgroup_get_live(cgrp);
5748 cgrp = ERR_PTR(-ENOTDIR);
5752 cgrp = ERR_PTR(-ENOENT);
5755 mutex_unlock(&cgroup_mutex);
5758 EXPORT_SYMBOL_GPL(cgroup_get_from_path);
5761 * cgroup_get_from_fd - get a cgroup pointer from a fd
5762 * @fd: fd obtained by open(cgroup2_dir)
5764 * Find the cgroup from a fd which should be obtained
5765 * by opening a cgroup directory. Returns a pointer to the
5766 * cgroup on success. ERR_PTR is returned if the cgroup
5769 struct cgroup *cgroup_get_from_fd(int fd)
5771 struct cgroup_subsys_state *css;
5772 struct cgroup *cgrp;
5777 return ERR_PTR(-EBADF);
5779 css = css_tryget_online_from_dir(f->f_path.dentry, NULL);
5782 return ERR_CAST(css);
5785 if (!cgroup_on_dfl(cgrp)) {
5787 return ERR_PTR(-EBADF);
5792 EXPORT_SYMBOL_GPL(cgroup_get_from_fd);
5795 * sock->sk_cgrp_data handling. For more info, see sock_cgroup_data
5796 * definition in cgroup-defs.h.
5798 #ifdef CONFIG_SOCK_CGROUP_DATA
5800 #if defined(CONFIG_CGROUP_NET_PRIO) || defined(CONFIG_CGROUP_NET_CLASSID)
5802 DEFINE_SPINLOCK(cgroup_sk_update_lock);
5803 static bool cgroup_sk_alloc_disabled __read_mostly;
5805 void cgroup_sk_alloc_disable(void)
5807 if (cgroup_sk_alloc_disabled)
5809 pr_info("cgroup: disabling cgroup2 socket matching due to net_prio or net_cls activation\n");
5810 cgroup_sk_alloc_disabled = true;
5815 #define cgroup_sk_alloc_disabled false
5819 void cgroup_sk_alloc(struct sock_cgroup_data *skcd)
5821 if (cgroup_sk_alloc_disabled) {
5822 skcd->no_refcnt = 1;
5826 /* Don't associate the sock with unrelated interrupted task's cgroup. */
5833 struct css_set *cset;
5835 cset = task_css_set(current);
5836 if (likely(cgroup_tryget(cset->dfl_cgrp))) {
5837 skcd->val = (unsigned long)cset->dfl_cgrp;
5846 void cgroup_sk_clone(struct sock_cgroup_data *skcd)
5848 /* Socket clone path */
5850 if (skcd->no_refcnt)
5853 * We might be cloning a socket which is left in an empty
5854 * cgroup and the cgroup might have already been rmdir'd.
5855 * Don't use cgroup_get_live().
5857 cgroup_get(sock_cgroup_ptr(skcd));
5861 void cgroup_sk_free(struct sock_cgroup_data *skcd)
5863 if (skcd->no_refcnt)
5866 cgroup_put(sock_cgroup_ptr(skcd));
5869 #endif /* CONFIG_SOCK_CGROUP_DATA */
5871 #ifdef CONFIG_CGROUP_BPF
5872 int cgroup_bpf_update(struct cgroup *cgrp, struct bpf_prog *prog,
5873 enum bpf_attach_type type, bool overridable)
5875 struct cgroup *parent = cgroup_parent(cgrp);
5878 mutex_lock(&cgroup_mutex);
5879 ret = __cgroup_bpf_update(cgrp, parent, prog, type, overridable);
5880 mutex_unlock(&cgroup_mutex);
5883 #endif /* CONFIG_CGROUP_BPF */